init

.gitignore

@@ -0,0 +1,38 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+dist/
+build/
+*.egg
+
+# Environment
+.env
+.venv/
+venv/
+
+# Database
+backend/data/
+
+# Uploads
+backend/uploads/
+
+# Exports
+backend/exports/
+
+# Frontend
+frontend/node_modules/
+frontend/dist/
+
+# IDE
+.vscode/
+.idea/
+*.swp
+*.swo
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Logs
+*.log

Dockerfile

@@ -0,0 +1,22 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# 安装系统依赖
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    gcc \
+    && rm -rf /var/lib/apt/lists/*
+
+# 安装 Python 依赖
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# 复制后端代码
+COPY app/ ./app/
+
+# 创建必要目录
+RUN mkdir -p /app/uploads/transcripts /app/uploads/persons /app/exports /app/data
+
+EXPOSE 8000
+
+CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "8000"]

backend/.env.example

@@ -0,0 +1,18 @@
+# LLM 配置
+LLM_PROVIDER=minimax
+LLM_MODEL=MiniMax-M2.7
+LLM_BASE_URL=https://api.minimax.chat/v1
+LLM_API_KEY=your-api-key-here
+
+# 提取参数
+SEGMENT_MAX_LINES=500
+STORY_MIN_LINES=50
+CONFIDENCE_THRESHOLD=0.5
+TEMPERATURE=0.3
+
+# 应用
+APP_HOST=0.0.0.0
+APP_PORT=8000
+UPLOAD_DIR=/app/uploads
+EXPORT_DIR=/app/exports
+DB_DIR=/app/data

backend/app/api/export.py

@@ -0,0 +1,195 @@
+import os
+import json
+import zipfile
+from typing import Annotated
+from fastapi import APIRouter, HTTPException, Depends
+from fastapi.responses import FileResponse
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select, text
+
+from app.database import get_db
+from app.models.story import Story
+from app.models.person import Person
+from app.config import settings
+from app.services.docx_generator import generate_person_doc
+
+router = APIRouter(prefix="/export", tags=["导出"])
+
+UNKNOWN_PERSON_ID = "unknown"
+UNKNOWN_PERSON_NAME = "暂无"
+
+
+@router.get("/list")
+async def export_list(db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取导出列表（所有已匹配的讲述者，含'暂无'）"""
+    # 获取所有已匹配的故事按 person_id 分组
+    result = await db.execute(
+        select(Story.person_id)
+        .where(Story.match_status == "matched")
+        .distinct()
+    )
+    person_ids = [row[0] for row in result.all()]
+
+    items = []
+    for pid in person_ids:
+        if pid == UNKNOWN_PERSON_ID:
+            story_result = await db.execute(
+                select(Story).where(
+                    Story.person_id == UNKNOWN_PERSON_ID,
+                    Story.match_status == "matched",
+                )
+            )
+            stories = story_result.scalars().all()
+            items.append({
+                "person_id": UNKNOWN_PERSON_ID,
+                "person_name": UNKNOWN_PERSON_NAME,
+                "story_count": len(stories),
+                "has_photo": False,
+            })
+        else:
+            person = await db.get(Person, pid)
+            if not person:
+                continue
+            story_result = await db.execute(
+                select(Story).where(
+                    Story.person_id == person.id,
+                    Story.match_status == "matched",
+                )
+            )
+            stories = story_result.scalars().all()
+            items.append({
+                "person_id": person.id,
+                "person_name": person.name,
+                "story_count": len(stories),
+                "has_photo": bool(person.photo_path),
+            })
+    return items
+
+
+@router.get("/preview/{person_id}")
+async def preview_export(person_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """预览某讲述者的文档内容"""
+    if person_id == UNKNOWN_PERSON_ID:
+        person_name = UNKNOWN_PERSON_NAME
+        person_info = ""
+        has_photo = False
+    else:
+        person = await db.get(Person, person_id)
+        if not person:
+            raise HTTPException(404, "讲述者不存在")
+        person_name = person.name
+        person_info = person.info
+        has_photo = bool(person.photo_path)
+
+    story_result = await db.execute(
+        select(Story).where(
+            Story.person_id == person_id,
+            Story.match_status == "matched",
+        )
+    )
+    stories = story_result.scalars().all()
+
+    return {
+        "person_name": person_name,
+        "person_info": person_info,
+        "has_photo": has_photo,
+        "stories": [{"title": s.title, "summary": s.summary, "tags": json.loads(s.tags) if s.tags else [], "content": s.content, "raw_material": s.raw_material or ""} for s in stories],
+    }
+
+
+@router.get("/download/{person_id}")
+async def download_one(person_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """下载单个讲述者的 DOCX"""
+    if person_id == UNKNOWN_PERSON_ID:
+        person_name = UNKNOWN_PERSON_NAME
+        person_info = ""
+        photo_path = ""
+    else:
+        person = await db.get(Person, person_id)
+        if not person:
+            raise HTTPException(404, "讲述者不存在")
+        person_name = person.name
+        person_info = person.info
+        photo_path = person.photo_path or ""
+
+    story_result = await db.execute(
+        select(Story).where(
+            Story.person_id == person_id,
+            Story.match_status == "matched",
+        )
+    )
+    stories = story_result.scalars().all()
+    if not stories:
+        raise HTTPException(400, "该讲述者没有已匹配的故事")
+
+    output_path = os.path.join(settings.EXPORT_DIR, f"{person_name}_故事集.docx")
+    generate_person_doc(
+        person_name=person_name,
+        person_info=person_info,
+        photo_path=photo_path,
+        stories=[{"title": s.title, "summary": s.summary, "tags": json.loads(s.tags) if s.tags else [], "content": s.content} for s in stories],
+        output_path=output_path,
+    )
+
+    return FileResponse(
+        output_path,
+        media_type="application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+        filename=f"{person_name}_故事集.docx",
+    )
+
+
+@router.get("/download-all")
+async def download_all(db: Annotated[AsyncSession, Depends(get_db)]):
+    """批量下载所有已匹配讲述者的文档（ZIP）"""
+    # 获取所有已匹配的 person_id
+    result = await db.execute(
+        select(Story.person_id)
+        .where(Story.match_status == "matched")
+        .distinct()
+    )
+    person_ids = [row[0] for row in result.all()]
+
+    if not person_ids:
+        raise HTTPException(400, "没有可导出的文档")
+
+    # 生成 ZIP
+    zip_path = os.path.join(settings.EXPORT_DIR, "故事集_全部.zip")
+    with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
+        for pid in person_ids:
+            if pid == UNKNOWN_PERSON_ID:
+                person_name = UNKNOWN_PERSON_NAME
+                person_info = ""
+                photo_path = ""
+            else:
+                person = await db.get(Person, pid)
+                if not person:
+                    continue
+                person_name = person.name
+                person_info = person.info
+                photo_path = person.photo_path or ""
+
+            story_result = await db.execute(
+                select(Story).where(
+                    Story.person_id == pid,
+                    Story.match_status == "matched",
+                )
+            )
+            stories = story_result.scalars().all()
+            if not stories:
+                continue
+
+            output_path = os.path.join(settings.EXPORT_DIR, f"{person_name}_故事集.docx")
+            generate_person_doc(
+                person_name=person_name,
+                person_info=person_info,
+                photo_path=photo_path,
+                stories=[{"title": s.title, "summary": s.summary, "tags": json.loads(s.tags) if s.tags else [], "content": s.content} for s in stories],
+                output_path=output_path,
+            )
+            zf.write(output_path, f"{person_name}_故事集.docx")
+
+    return FileResponse(
+        zip_path,
+        media_type="application/zip",
+        filename="故事集_全部.zip",
+    )

backend/app/api/extraction.py

@@ -0,0 +1,112 @@
+import asyncio
+import json
+from typing import Annotated, Optional
+from fastapi import APIRouter, HTTPException, Depends, Query
+from fastapi.responses import StreamingResponse
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select
+
+from app.database import get_db
+from app.models.transcript import Transcript
+from app.state import task_manager
+from app.services.extractor import run_extraction
+
+router = APIRouter(prefix="/extraction", tags=["故事提取"])
+
+
+@router.post("/start")
+async def start_extraction(
+    task_id: Optional[str] = Query(None),
+    db: Annotated[AsyncSession, Depends(get_db)] = None,
+):
+    """启动提取任务，支持多用户并发"""
+    # 如果传了 task_id，复用已有任务
+    if task_id:
+        state = task_manager.get_task(task_id)
+        if state and state.is_running:
+            return {"task_id": task_id, "message": "任务正在进行中"}
+    else:
+        state = None
+
+    # 检查是否有待处理的文字稿
+    result = await db.execute(select(Transcript).where(Transcript.status == "pending"))
+    pending = result.scalars().all()
+
+    if not pending:
+        # 没有 pending 的，把所有非 pending 的文件重置为 pending（包括 completed/processing/error）
+        result2 = await db.execute(
+            select(Transcript).where(Transcript.status != "pending")
+        )
+        others = result2.scalars().all()
+        if others:
+            for t in others:
+                t.status = "pending"
+                t.error_message = ""
+            await db.commit()
+            pending = others
+        else:
+            raise HTTPException(400, "没有可提取的文字稿，请先上传文件")
+
+    total = len(pending)
+
+    # 创建新任务
+    state = task_manager.create_task()
+    state.reset()
+    state.is_running = True
+
+    asyncio.create_task(run_extraction(state))
+
+    return {"task_id": state.task_id, "total_files": total}
+
+
+@router.get("/status")
+async def get_status(task_id: Optional[str] = Query(None)):
+    """获取提取进度"""
+    if task_id:
+        state = task_manager.get_task(task_id)
+        if state:
+            return {**state.progress, "task_id": state.task_id}
+    return {"status": "idle", "percent": 0}
+
+
+@router.get("/status/stream")
+async def stream_status(task_id: Optional[str] = Query(None)):
+    """SSE 实时进度推送（支持多用户）"""
+    if not task_id:
+        raise HTTPException(400, "缺少 task_id 参数")
+
+    state = task_manager.get_task(task_id)
+    if not state:
+        raise HTTPException(404, "任务不存在")
+
+    async def event_generator():
+        # 立即推送当前状态
+        yield f"event: progress\ndata: {json.dumps({**state.progress, 'task_id': state.task_id}, ensure_ascii=False)}\n\n"
+        if state.progress.get("status") in ("completed", "error", "stopped"):
+            return
+        while True:
+            data = await state.get_update()
+            yield f"event: progress\ndata: {json.dumps({**data, 'task_id': state.task_id}, ensure_ascii=False)}\n\n"
+            if data.get("status") in ("completed", "error", "stopped"):
+                break
+
+    return StreamingResponse(
+        event_generator(),
+        media_type="text/event-stream",
+        headers={"Cache-Control": "no-cache", "X-Accel-Buffering": "no"},
+    )
+
+
+@router.post("/stop")
+async def stop_extraction(task_id: Optional[str] = Query(None)):
+    """停止提取任务"""
+    if task_id:
+        state = task_manager.get_task(task_id)
+    else:
+        state = None
+
+    if not state or not state.is_running:
+        raise HTTPException(400, "没有正在进行的提取任务")
+
+    state.cancel_event.set()
+    return {"message": "正在停止...", "task_id": state.task_id}

backend/app/api/files.py

@@ -0,0 +1,136 @@
+import os
+import uuid
+from typing import Annotated
+from fastapi import APIRouter, UploadFile, File, HTTPException, Depends
+from fastapi.responses import FileResponse
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select, delete
+
+from app.database import get_db
+from app.models.transcript import Transcript
+from app.models.person import Person
+from app.schemas.schemas import TranscriptOut, PersonOut
+from app.services.file_service import file_service
+from app.services.docx_parser import parse_transcript, parse_person_info
+
+router = APIRouter(prefix="/files", tags=["文件管理"])
+
+
+# ===== 文字稿 =====
+
+@router.post("/transcripts", response_model=list[TranscriptOut])
+async def upload_transcripts(files: list[UploadFile], db: Annotated[AsyncSession, Depends(get_db)]):
+    """批量上传文字稿"""
+    results = []
+    for file in files:
+        if not file.filename.endswith(".docx"):
+            raise HTTPException(400, f"仅支持 .docx 格式: {file.filename}")
+
+        saved = await file_service.save_transcript(file)
+
+        # 解析行数
+        paragraphs = parse_transcript(saved["file_path"])
+        line_count = len(paragraphs)
+
+        record = Transcript(
+            id=saved["id"],
+            filename=saved["filename"],
+            file_path=saved["file_path"],
+            line_count=line_count,
+            file_size=saved["file_size"],
+            status="pending",
+        )
+        db.add(record)
+        results.append(record)
+
+    await db.commit()
+    for r in results:
+        await db.refresh(r)
+    return results
+
+
+@router.get("/transcripts", response_model=list[TranscriptOut])
+async def list_transcripts(db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取文字稿列表"""
+    result = await db.execute(select(Transcript).order_by(Transcript.uploaded_at.desc()))
+    return result.scalars().all()
+
+
+@router.delete("/transcripts/{transcript_id}")
+async def delete_transcript(transcript_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """删除文字稿"""
+    record = await db.get(Transcript, transcript_id)
+    if not record:
+        raise HTTPException(404, "文字稿不存在")
+    file_service.delete_file(record.file_path)
+    await db.delete(record)
+    await db.commit()
+    return {"message": "已删除"}
+
+
+# ===== 讲述者信息 =====
+
+@router.post("/persons", response_model=list[PersonOut])
+async def upload_persons(files: list[UploadFile], db: Annotated[AsyncSession, Depends(get_db)]):
+    """批量上传讲述者信息"""
+    results = []
+    for file in files:
+        if not file.filename.endswith(".docx"):
+            raise HTTPException(400, f"仅支持 .docx 格式: {file.filename}")
+
+        saved = await file_service.save_person(file)
+
+        # 解析个人信息
+        info = parse_person_info(saved["file_path"])
+
+        record = Person(
+            id=saved["id"],
+            name=info["name"],
+            filename=saved["filename"],
+            file_path=saved["file_path"],
+            photo_path=info["photos"][0] if info["photos"] else "",
+            info=info["info_text"],
+        )
+        db.add(record)
+        results.append(record)
+
+    await db.commit()
+    for r in results:
+        await db.refresh(r)
+    return results
+
+
+@router.get("/persons", response_model=list[PersonOut])
+async def list_persons(db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取讲述者列表"""
+    result = await db.execute(select(Person).order_by(Person.uploaded_at.desc()))
+    return result.scalars().all()
+
+
+@router.delete("/persons/{person_id}")
+async def delete_person(person_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """删除讲述者信息"""
+    record = await db.get(Person, person_id)
+    if not record:
+        raise HTTPException(404, "讲述者不存在")
+    file_service.delete_file(record.file_path)
+    if record.photo_path:
+        file_service.delete_file(record.photo_path)
+    await db.delete(record)
+    await db.commit()
+    return {"message": "已删除"}
+
+
+@router.get("/persons/{person_id}/preview")
+async def preview_person(person_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """预览讲述者信息"""
+    record = await db.get(Person, person_id)
+    if not record:
+        raise HTTPException(404, "讲述者不存在")
+    return {
+        "id": record.id,
+        "name": record.name,
+        "nickname": record.nickname,
+        "info": record.info,
+        "photo_path": record.photo_path,
+    }

backend/app/api/matching.py

@@ -0,0 +1,76 @@
+from typing import Annotated
+from fastapi import APIRouter, HTTPException, Depends
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select
+
+from app.database import get_db
+from app.models.story import Story
+from app.models.person import Person
+from app.schemas.schemas import MatchCreate, MatchOut
+
+router = APIRouter(prefix="/matches", tags=["匹配管理"])
+
+# 特殊讲述者 ID：暂无
+UNKNOWN_PERSON_ID = "unknown"
+UNKNOWN_PERSON_NAME = "暂无"
+
+
+@router.post("")
+async def create_match(data: MatchCreate, db: Annotated[AsyncSession, Depends(get_db)]):
+    """创建匹配关系（支持 person_id='unknown' 表示暂无讲述者）"""
+    story = await db.get(Story, data.story_id)
+    if not story or story.match_status == "deleted":
+        raise HTTPException(404, "故事不存在")
+
+    if data.person_id == UNKNOWN_PERSON_ID:
+        person_name = UNKNOWN_PERSON_NAME
+    else:
+        person = await db.get(Person, data.person_id)
+        if not person:
+            raise HTTPException(404, "讲述者不存在")
+        person_name = person.name
+
+    story.person_id = data.person_id
+    story.match_status = "matched"
+    await db.commit()
+    await db.refresh(story)
+
+    return {"message": "匹配成功", "story_id": story.id, "person_id": data.person_id, "person_name": person_name}
+
+
+@router.delete("/{story_id}")
+async def remove_match(story_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """取消匹配"""
+    story = await db.get(Story, story_id)
+    if not story:
+        raise HTTPException(404, "故事不存在")
+
+    story.person_id = None
+    story.match_status = "pending"
+    await db.commit()
+    return {"message": "已取消匹配"}
+
+
+@router.get("", response_model=list[MatchOut])
+async def list_matches(db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取所有匹配关系"""
+    result = await db.execute(
+        select(Story, Person)
+        .join(Person, Story.person_id == Person.id, isouter=True)
+        .where(Story.match_status == "matched")
+    )
+    matches = []
+    for story, person in result.all():
+        if story.person_id == UNKNOWN_PERSON_ID:
+            person_name = UNKNOWN_PERSON_NAME
+        elif person:
+            person_name = person.name
+        else:
+            person_name = UNKNOWN_PERSON_NAME
+        matches.append({
+            "story_id": story.id,
+            "person_id": story.person_id,
+            "person_name": person_name,
+            "story_title": story.title,
+        })
+    return matches

backend/app/api/settings.py

@@ -0,0 +1,70 @@
+from typing import Annotated
+from fastapi import APIRouter, Depends
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select
+
+from app.database import get_db
+from app.models.config import Config
+from app.schemas.schemas import SettingsOut, SettingsUpdate
+from app.services.llm_client import llm_client
+from app.config import settings
+
+router = APIRouter(prefix="/settings", tags=["系统设置"])
+
+
+async def _get_config_value(db: AsyncSession, key: str, default: str = "") -> str:
+    """获取配置值"""
+    result = await db.execute(select(Config).where(Config.key == key))
+    config = result.scalar_one_or_none()
+    return config.value if config else default
+
+
+@router.get("", response_model=SettingsOut)
+async def get_settings(db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取系统配置"""
+    return SettingsOut(
+        llm_provider=await _get_config_value(db, "llm_provider", settings.LLM_PROVIDER),
+        llm_model=await _get_config_value(db, "llm_model", settings.LLM_MODEL),
+        llm_base_url=await _get_config_value(db, "llm_base_url", settings.LLM_BASE_URL),
+        llm_api_key=await _get_config_value(db, "llm_api_key", "****"),
+        segment_max_lines=int(await _get_config_value(db, "segment_max_lines", str(settings.SEGMENT_MAX_LINES))),
+        story_min_lines=int(await _get_config_value(db, "story_min_lines", str(settings.STORY_MIN_LINES))),
+        confidence_threshold=float(await _get_config_value(db, "confidence_threshold", str(settings.CONFIDENCE_THRESHOLD))),
+        temperature=float(await _get_config_value(db, "temperature", str(settings.TEMPERATURE))),
+    )
+
+
+@router.put("")
+async def update_settings(data: SettingsUpdate, db: Annotated[AsyncSession, Depends(get_db)]):
+    """更新系统配置"""
+    updates = data.model_dump(exclude_none=True)
+    for key, value in updates.items():
+        # API Key 脱敏处理
+        if key == "llm_api_key" and value == "****":
+            continue
+
+        existing = await db.execute(select(Config).where(Config.key == key))
+        config = existing.scalar_one_or_none()
+        if config:
+            config.value = str(value)
+        else:
+            db.add(Config(key=key, value=str(value)))
+
+    await db.commit()
+
+    # 更新 LLM 客户端配置
+    if "llm_base_url" in updates:
+        llm_client.update_config(base_url=updates["llm_base_url"])
+    if "llm_api_key" in updates:
+        llm_client.update_config(api_key=updates["llm_api_key"])
+    if "llm_model" in updates:
+        llm_client.update_config(model=updates["llm_model"])
+
+    return {"message": "配置已保存"}
+
+
+@router.post("/test-llm")
+async def test_llm():
+    """测试 LLM 连接"""
+    result = await llm_client.test_connection()
+    return result

backend/app/api/stories.py

@@ -0,0 +1,64 @@
+from typing import Annotated
+from fastapi import APIRouter, HTTPException, Depends
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select, update
+
+from app.database import get_db
+from app.models.story import Story
+from app.schemas.schemas import StoryOut, StoryEdit
+
+router = APIRouter(prefix="/stories", tags=["故事管理"])
+
+
+@router.get("", response_model=list[StoryOut])
+async def list_stories(db: Annotated[AsyncSession, Depends(get_db)], status: str = "all"):
+    """获取故事列表，支持筛选：all/matched/pending"""
+    query = select(Story).where(Story.match_status != "deleted")
+    if status == "matched":
+        query = query.where(Story.match_status == "matched")
+    elif status == "pending":
+        query = query.where(Story.match_status == "pending")
+    query = query.order_by(Story.created_at.desc())
+    result = await db.execute(query)
+    return result.scalars().all()
+
+
+@router.get("/{story_id}", response_model=StoryOut)
+async def get_story(story_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """获取故事详情"""
+    story = await db.get(Story, story_id)
+    if not story or story.match_status == "deleted":
+        raise HTTPException(404, "故事不存在")
+    return story
+
+
+@router.put("/{story_id}", response_model=StoryOut)
+async def edit_story(story_id: str, data: StoryEdit, db: Annotated[AsyncSession, Depends(get_db)]):
+    """编辑故事"""
+    story = await db.get(Story, story_id)
+    if not story or story.match_status == "deleted":
+        raise HTTPException(404, "故事不存在")
+
+    if data.title is not None:
+        story.title = data.title
+    if data.summary is not None:
+        story.summary = data.summary
+    if data.content is not None:
+        story.content = data.content
+
+    await db.commit()
+    await db.refresh(story)
+    return story
+
+
+@router.delete("/{story_id}")
+async def delete_story(story_id: str, db: Annotated[AsyncSession, Depends(get_db)]):
+    """删除故事（标记为 deleted）"""
+    story = await db.get(Story, story_id)
+    if not story or story.match_status == "deleted":
+        raise HTTPException(404, "故事不存在")
+
+    story.match_status = "deleted"
+    story.person_id = None
+    await db.commit()
+    return {"message": "已删除"}

backend/app/config.py

@@ -0,0 +1,47 @@
+import os
+import sys
+from dotenv import load_dotenv
+
+load_dotenv()
+
+
+def _get_base_dir() -> str:
+    """获取基础目录（支持 PyInstaller 打包后的路径）"""
+    if getattr(sys, 'frozen', False):
+        # PyInstaller 打包后，exe 所在目录
+        return os.path.dirname(sys.executable)
+    else:
+        # 开发模式，backend 目录
+        return os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+
+
+BASE_DIR = _get_base_dir()
+
+
+class Settings:
+    # App
+    APP_HOST: str = os.getenv("APP_HOST", "0.0.0.0")
+    APP_PORT: int = int(os.getenv("APP_PORT", "8000"))
+    UPLOAD_DIR: str = os.getenv("UPLOAD_DIR", os.path.join(BASE_DIR, "uploads"))
+    EXPORT_DIR: str = os.getenv("EXPORT_DIR", os.path.join(BASE_DIR, "exports"))
+    DB_DIR: str = os.getenv("DB_DIR", os.path.join(BASE_DIR, "data"))
+
+    # LLM
+    LLM_PROVIDER: str = os.getenv("LLM_PROVIDER", "xiaomi")
+    LLM_MODEL: str = os.getenv("LLM_MODEL", "MiMo-V2-Flash")
+    LLM_BASE_URL: str = os.getenv("LLM_BASE_URL", "https://api.xiaomimimo.com/v1")
+    LLM_API_KEY: str = os.getenv("LLM_API_KEY", "")
+
+    # Extraction
+    SEGMENT_MAX_LINES: int = int(os.getenv("SEGMENT_MAX_LINES", "500"))
+    STORY_MIN_LINES: int = int(os.getenv("STORY_MIN_LINES", "50"))
+    CONFIDENCE_THRESHOLD: float = float(os.getenv("CONFIDENCE_THRESHOLD", "0.5"))
+    TEMPERATURE: float = float(os.getenv("TEMPERATURE", "0.3"))
+
+    @property
+    def database_url(self) -> str:
+        os.makedirs(self.DB_DIR, exist_ok=True)
+        return f"sqlite+aiosqlite:///{self.DB_DIR}/story_extractor.db"
+
+
+settings = Settings()

backend/app/database.py

@@ -0,0 +1,26 @@
+from typing import AsyncGenerator
+from sqlalchemy.ext.asyncio import create_async_engine, AsyncSession, async_sessionmaker
+from sqlalchemy.orm import DeclarativeBase
+from sqlalchemy import text
+
+from app.config import settings
+
+engine = create_async_engine(settings.database_url, echo=False, connect_args={"check_same_thread": False})
+async_session = async_sessionmaker(engine, class_=AsyncSession, expire_on_commit=False)
+
+
+class Base(DeclarativeBase):
+    pass
+
+
+async def init_db():
+    async with engine.begin() as conn:
+        await conn.run_sync(Base.metadata.create_all)
+        # 启用 WAL 模式，提升并发读写性能
+        await conn.execute(text("PRAGMA journal_mode=WAL"))
+        await conn.execute(text("PRAGMA busy_timeout=5000"))
+
+
+async def get_db() -> AsyncGenerator[AsyncSession, None]:
+    async with async_session() as session:
+        yield session

backend/app/main.py

@@ -0,0 +1,64 @@
+import os
+import sys
+import logging
+from contextlib import asynccontextmanager
+from fastapi import FastAPI
+from fastapi.staticfiles import StaticFiles
+from fastapi.middleware.cors import CORSMiddleware
+
+from app.config import settings, BASE_DIR
+from app.database import init_db
+from app.api import files, extraction, stories, matching, export, settings as settings_api
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+
+def _get_frontend_dir() -> str:
+    """获取前端静态文件目录（支持 PyInstaller 打包）"""
+    if getattr(sys, 'frozen', False):
+        # 打包后：exe 同级目录下的 frontend/dist
+        return os.path.join(BASE_DIR, "frontend", "dist")
+    else:
+        # 开发模式：项目根目录下的 frontend/dist
+        return os.path.join(os.path.dirname(BASE_DIR), "frontend", "dist")
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    """应用生命周期：启动时初始化数据库"""
+    logger.info("初始化数据库...")
+    await init_db()
+    logger.info("数据库初始化完成")
+    yield
+    logger.info("应用关闭")
+
+
+app = FastAPI(
+    title="Story Extractor",
+    description="直播文字稿故事提取工具",
+    version="1.0.0",
+    lifespan=lifespan,
+)
+
+# CORS
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# API 路由
+app.include_router(files.router, prefix="/api/v1")
+app.include_router(extraction.router, prefix="/api/v1")
+app.include_router(stories.router, prefix="/api/v1")
+app.include_router(matching.router, prefix="/api/v1")
+app.include_router(export.router, prefix="/api/v1")
+app.include_router(settings_api.router, prefix="/api/v1")
+
+# 托管前端静态文件（放在最后，作为 fallback）
+frontend_dist = _get_frontend_dir()
+if os.path.exists(frontend_dist):
+    app.mount("/", StaticFiles(directory=frontend_dist, html=True), name="static")

backend/app/models/config.py

@@ -0,0 +1,11 @@
+from sqlalchemy import Column, String, Text
+from sqlalchemy.sql import func
+
+from app.database import Base
+
+
+class Config(Base):
+    __tablename__ = "config"
+
+    key = Column(String, primary_key=True)
+    value = Column(Text, nullable=False)

backend/app/models/person.py

@@ -0,0 +1,17 @@
+from sqlalchemy import Column, String, Text, DateTime
+from sqlalchemy.sql import func
+
+from app.database import Base
+
+
+class Person(Base):
+    __tablename__ = "persons"
+
+    id = Column(String, primary_key=True)
+    name = Column(String, nullable=False)
+    nickname = Column(String, default="")  # 文字稿中的昵称
+    filename = Column(String, nullable=False)
+    file_path = Column(String, nullable=False)
+    photo_path = Column(String, default="")
+    info = Column(Text, default="")  # JSON
+    uploaded_at = Column(DateTime, server_default=func.now())

backend/app/models/story.py

@@ -0,0 +1,26 @@
+import uuid
+
+from sqlalchemy import Column, String, Integer, Float, Text, DateTime, ForeignKey
+from sqlalchemy.sql import func
+
+from app.database import Base
+
+
+class Story(Base):
+    __tablename__ = "stories"
+
+    id = Column(String, primary_key=True, default=lambda: uuid.uuid4().hex)
+    title = Column(String, nullable=False)
+    summary = Column(Text, default="")
+    tags = Column(Text, default="")  # JSON: ["标签1", "标签2", ...]
+    content = Column(Text, default="")
+    raw_material = Column(Text, default="")  # JSON: 第四步重组后的素材
+    speaker_nickname = Column(String, default="")
+    source_transcript_id = Column(String, ForeignKey("transcripts.id"), nullable=False)
+    source_lines = Column(Text, default="")  # JSON: {start, end}
+    duration_minutes = Column(Float, default=0)
+    confidence = Column(Float, default=0)
+    confidence_level = Column(String, default="medium")  # high/medium/low
+    person_id = Column(String, ForeignKey("persons.id"), nullable=True)
+    match_status = Column(String, default="pending")  # pending/matched/deleted
+    created_at = Column(DateTime, server_default=func.now())

backend/app/models/transcript.py

@@ -0,0 +1,17 @@
+from sqlalchemy import Column, String, Integer, Float, Text, DateTime
+from sqlalchemy.sql import func
+
+from app.database import Base
+
+
+class Transcript(Base):
+    __tablename__ = "transcripts"
+
+    id = Column(String, primary_key=True)
+    filename = Column(String, nullable=False)
+    file_path = Column(String, nullable=False)
+    line_count = Column(Integer, default=0)
+    file_size = Column(Integer, default=0)
+    status = Column(String, default="pending")  # pending/processing/done/error
+    error_message = Column(Text, default="")
+    uploaded_at = Column(DateTime, server_default=func.now())

backend/app/schemas/schemas.py

@@ -0,0 +1,100 @@
+from pydantic import BaseModel
+from datetime import datetime
+
+
+class TranscriptOut(BaseModel):
+    id: str
+    filename: str
+    file_path: str
+    line_count: int
+    file_size: int
+    status: str
+    error_message: str
+    uploaded_at: datetime
+
+    class Config:
+        from_attributes = True
+
+
+class PersonOut(BaseModel):
+    id: str
+    name: str
+    nickname: str
+    filename: str
+    file_path: str
+    photo_path: str
+    info: str
+    uploaded_at: datetime
+
+    class Config:
+        from_attributes = True
+
+
+class StoryOut(BaseModel):
+    id: str
+    title: str
+    summary: str
+    content: str
+    raw_material: str
+    speaker_nickname: str
+    source_transcript_id: str
+    source_lines: str
+    duration_minutes: float
+    confidence: float
+    confidence_level: str
+    person_id: str | None
+    match_status: str
+    created_at: datetime
+
+    class Config:
+        from_attributes = True
+
+
+class StoryEdit(BaseModel):
+    title: str | None = None
+    summary: str | None = None
+    content: str | None = None
+
+
+class MatchCreate(BaseModel):
+    story_id: str
+    person_id: str
+
+
+class MatchOut(BaseModel):
+    story_id: str
+    person_id: str
+    person_name: str
+    story_title: str
+
+    class Config:
+        from_attributes = True
+
+
+class ExportItemOut(BaseModel):
+    person_id: str
+    person_name: str
+    story_count: int
+    has_photo: bool
+
+
+class SettingsOut(BaseModel):
+    llm_provider: str
+    llm_model: str
+    llm_base_url: str
+    llm_api_key: str
+    segment_max_lines: int
+    story_min_lines: int
+    confidence_threshold: float
+    temperature: float
+
+
+class SettingsUpdate(BaseModel):
+    llm_provider: str | None = None
+    llm_model: str | None = None
+    llm_base_url: str | None = None
+    llm_api_key: str | None = None
+    segment_max_lines: int | None = None
+    story_min_lines: int | None = None
+    confidence_threshold: float | None = None
+    temperature: float | None = None

backend/app/services/docx_generator.py

@@ -0,0 +1,166 @@
+import os
+import re
+from docx import Document
+from docx.shared import Pt, Inches, Cm, RGBColor
+from docx.enum.text import WD_ALIGN_PARAGRAPH
+from docx.oxml.ns import qn
+
+
+def generate_person_doc(person_name: str, person_info: str, photo_path: str,
+                        stories: list[dict], output_path: str):
+    """
+    为一位讲述者生成最终 DOCX 文档
+
+    person_info: 个人信息文本
+    photo_path: 照片路径（可为空）
+    stories: [{title, content}]
+    """
+    doc = Document()
+
+    # 设置默认字体
+    style = doc.styles['Normal']
+    font = style.font
+    font.name = '宋体'
+    font.size = Pt(12)
+    style.element.rPr.rFonts.set(qn('w:eastAsia'), '宋体')
+
+    # 照片
+    if photo_path and os.path.exists(photo_path):
+        try:
+            doc.add_picture(photo_path, width=Inches(1.5))
+            last_paragraph = doc.paragraphs[-1]
+            last_paragraph.alignment = WD_ALIGN_PARAGRAPH.CENTER
+        except Exception:
+            pass
+
+    # 姓名
+    name_para = doc.add_paragraph()
+    name_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
+    name_run = name_para.add_run(person_name)
+    name_run.font.size = Pt(18)
+    name_run.font.bold = True
+    name_run.font.name = '黑体'
+    name_run.element.rPr.rFonts.set(qn('w:eastAsia'), '黑体')
+
+    # 个人信息
+    if person_info:
+        info_para = doc.add_paragraph()
+        info_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
+        info_run = info_para.add_run(person_info.strip())
+        info_run.font.size = Pt(11)
+        info_run.font.color.rgb = RGBColor(0x66, 0x66, 0x66)
+
+    # 分隔线
+    doc.add_paragraph("─" * 30).alignment = WD_ALIGN_PARAGRAPH.CENTER
+
+    # 故事
+    for i, story in enumerate(stories):
+        # 故事标题
+        title_para = doc.add_paragraph()
+        title_run = title_para.add_run(story.get("title", f"故事 {i + 1}"))
+        title_run.font.size = Pt(14)
+        title_run.font.bold = True
+        title_run.font.name = '黑体'
+        title_run.element.rPr.rFonts.set(qn('w:eastAsia'), '黑体')
+
+        # 故事梗概
+        summary = story.get("summary", "")
+        if summary:
+            p = doc.add_paragraph()
+            run = p.add_run("【故事梗概】")
+            run.font.size = Pt(11)
+            run.font.bold = True
+            run.font.color.rgb = RGBColor(0x33, 0x33, 0x99)
+            p = doc.add_paragraph(summary.strip())
+            p.paragraph_format.first_line_indent = Cm(0.74)
+            p.paragraph_format.line_spacing = 1.5
+            for run in p.runs:
+                run.font.size = Pt(11)
+
+        # 关键词标签
+        tags = story.get("tags", [])
+        if tags:
+            p = doc.add_paragraph()
+            run = p.add_run("【关键词标签】")
+            run.font.size = Pt(11)
+            run.font.bold = True
+            run.font.color.rgb = RGBColor(0x33, 0x33, 0x99)
+            tag_text = "、".join(tags)
+            p = doc.add_paragraph(tag_text)
+            p.paragraph_format.line_spacing = 1.5
+            for run in p.runs:
+                run.font.size = Pt(10)
+                run.font.color.rgb = RGBColor(0x66, 0x66, 0x66)
+
+        # 故事内容（支持 Markdown 格式）
+        content = story.get("content", "")
+        if content:
+            _add_markdown_content(doc, content)
+
+        # 多个故事之间加分隔
+        if i < len(stories) - 1:
+            doc.add_paragraph("─" * 30).alignment = WD_ALIGN_PARAGRAPH.CENTER
+
+    # 保存
+    os.makedirs(os.path.dirname(output_path), exist_ok=True)
+    doc.save(output_path)
+    return output_path
+
+
+def _add_markdown_content(doc, content: str):
+    """将 Markdown 格式的内容添加到 DOCX，支持标题、引用、列表等"""
+    lines = content.split("\n")
+    i = 0
+    while i < len(lines):
+        line = lines[i].strip()
+
+        if not line:
+            i += 1
+            continue
+
+        # ## 二级标题
+        if line.startswith("## "):
+            title = line[3:].strip()
+            p = doc.add_paragraph()
+            run = p.add_run(title)
+            run.font.size = Pt(13)
+            run.font.bold = True
+            run.font.name = '黑体'
+            run.element.rPr.rFonts.set(qn('w:eastAsia'), '黑体')
+            p.paragraph_format.space_before = Pt(12)
+
+        # ### 三级标题
+        elif line.startswith("### "):
+            title = line[4:].strip()
+            p = doc.add_paragraph()
+            run = p.add_run(title)
+            run.font.size = Pt(12)
+            run.font.bold = True
+            run.font.name = '黑体'
+            run.element.rPr.rFonts.set(qn('w:eastAsia'), '黑体')
+            p.paragraph_format.space_before = Pt(6)
+
+        # > 引用
+        elif line.startswith("> "):
+            quote_text = line[2:].strip()
+            p = doc.add_paragraph()
+            p.paragraph_format.left_indent = Cm(1)
+            run = p.add_run(quote_text)
+            run.font.italic = True
+            run.font.color.rgb = RGBColor(0x55, 0x55, 0x55)
+            run.font.size = Pt(11)
+
+        # - 列表项
+        elif line.startswith("- "):
+            item_text = line[2:].strip()
+            p = doc.add_paragraph(style='List Bullet')
+            run = p.add_run(item_text)
+            run.font.size = Pt(12)
+
+        # 普通段落
+        else:
+            p = doc.add_paragraph(line)
+            p.paragraph_format.first_line_indent = Cm(0.74)
+            p.paragraph_format.line_spacing = 1.5
+
+        i += 1

backend/app/services/docx_parser.py

@@ -0,0 +1,93 @@
+import re
+from docx import Document
+
+
+def parse_transcript(file_path: str) -> list[dict]:
+    """
+    解析直播回放文字稿 DOCX
+    格式：发言者(时间戳): 内容
+
+    返回：[{index, speaker, timestamp, content, raw_text}]
+    """
+    doc = Document(file_path)
+    paragraphs = []
+    for i, para in enumerate(doc.paragraphs):
+        text = para.text.strip()
+        if not text:
+            continue
+
+        # 解析格式：发言者(时间戳): 内容
+        match = re.match(r'^(.+?)\((\d{2}:\d{2}:\d{2})\)\s*[:：]\s*(.*)', text)
+        if match:
+            speaker = match.group(1).strip()
+            timestamp = match.group(2)
+            content = match.group(3).strip()
+        else:
+            # 尝试格式：发言者: 内容
+            match2 = re.match(r'^(.+?)\s*[:：]\s*(.*)', text)
+            if match2:
+                speaker = match2.group(1).strip()
+                timestamp = ""
+                content = match2.group(2).strip()
+            else:
+                speaker = ""
+                timestamp = ""
+                content = text
+
+        paragraphs.append({
+            "index": i,
+            "speaker": speaker,
+            "timestamp": timestamp,
+            "content": content,
+            "raw_text": text,
+        })
+
+    return paragraphs
+
+
+def is_teacher(speaker: str) -> bool:
+    """判断是否为老师/助教发言"""
+    teacher_keywords = ["卢慧老师", "静静", "督导", "老师"]
+    return any(kw in speaker for kw in teacher_keywords)
+
+
+def parse_person_info(file_path: str) -> dict:
+    """
+    解析讲述者个人信息 DOCX
+    返回：{name, info_text, photos: [path]}
+    """
+    doc = Document(file_path)
+    texts = []
+    photos = []
+
+    # 提取文本
+    for para in doc.paragraphs:
+        text = para.text.strip()
+        if text:
+            texts.append(text)
+
+    # 提取图片
+    import os
+    photo_dir = os.path.join(os.path.dirname(file_path), "photos")
+    os.makedirs(photo_dir, exist_ok=True)
+
+    for i, rel in enumerate(doc.part.rels.values()):
+        if "image" in rel.reltype:
+            image = rel.target_part
+            photo_name = f"{os.path.splitext(os.path.basename(file_path))[0]}_photo_{i}.{image.content_type.split('/')[-1]}"
+            photo_path = os.path.join(photo_dir, photo_name)
+            with open(photo_path, "wb") as f:
+                f.write(image.blob)
+            photos.append(photo_path)
+
+    # 从文件名提取姓名
+    filename = os.path.basename(file_path)
+    name = re.sub(r'[-_].*$', '', filename).strip()
+    # 去掉扩展名
+    name = os.path.splitext(name)[0].strip()
+
+    return {
+        "name": name,
+        "info_text": "\n".join(texts),
+        "photos": photos,
+    }

backend/app/services/extractor.py

@@ -0,0 +1,957 @@
+"""
+故事提取器 - v4 五步流水线策略
+第一步：代码预处理（识别老师→过滤杂音→时间范围截取→过滤短段落/非主讲述人）
+第二步：判断故事 + 提取事实（短段落直接提取，长段落先判断再切片提取）
+第三步：切片提取事实（按自然段切分，并行提取）
+第四步：重组（纯代码，事件/情感/原话/人物去重）
+第五步：生成最终故事（基于素材生成5部分结构化故事）
+"""
+import json
+import asyncio
+import logging
+
+from sqlalchemy import select
+
+from app.database import async_session
+from app.models.story import Story
+from app.models.transcript import Transcript
+from app.services.docx_parser import parse_transcript
+from app.services.preprocessor import Preprocessor
+from app.services.llm_client import llm_client
+from app.state import TaskState
+
+logger = logging.getLogger(__name__)
+
+# 并发控制
+MAX_CONCURRENT_FILES = 3
+MAX_CONCURRENT_SEGMENTS = 5
+
+# 短段落阈值（字数）：<=4000 字直接提取事实，>4000 字先判断是否有故事
+SHORT_THRESHOLD = 4000
+
+# 切片参数
+CHUNK_MAX_CHARS = 3000
+CHUNK_OVERLAP_PARAGRAPHS = 5
+
+
+# ============================================================
+# 主入口
+# ============================================================
+async def run_extraction(state: TaskState):
+    """主提取流程：文件间串行（避免 API 限流），文件内各步骤并行"""
+    state.is_running = True
+    try:
+        async with async_session() as db:
+            result = await db.execute(
+                select(Transcript).where(Transcript.status == "pending")
+            )
+            transcripts = result.scalars().all()
+
+        if not transcripts:
+            state.update(status="completed", percent=100)
+            return
+
+        file_total = len(transcripts)
+        state.update(
+            status="running", percent=0,
+            files_total=file_total, files_completed=0,
+            stories_found=0,
+            current_file="", action=f"共发现 {file_total} 份文字稿，开始处理...",
+            phase="preprocess", phase_percent=0, phase_detail="准备中",
+        )
+
+        total_stories = [0]
+
+        # 文件间串行处理（避免 API 限流）
+        for file_index, transcript in enumerate(transcripts):
+            await _process_file(state, file_index, file_total, transcript, total_stories)
+
+        state.update(
+            status="completed", percent=100,
+            files_completed=file_total,
+            stories_found=total_stories[0],
+            current_file="", action="全部处理完成！",
+            phase="preprocess", phase_percent=100, phase_detail="全部完成",
+        )
+        state.is_running = False
+
+    except asyncio.CancelledError:
+        state.update(status="stopped", action="提取已停止")
+        state.is_running = False
+    except Exception as e:
+        logger.exception("Extraction failed")
+        state.update(status="error", action=str(e))
+        state.is_running = False
+
+
+# ============================================================
+# 单文件处理（五步流水线）
+# ============================================================
+async def _process_file(state: TaskState, file_index: int, file_total: int,
+                       transcript: Transcript, total_stories: list):
+    """处理单个文件 - 五步流水线"""
+    try:
+        async with async_session() as db:
+            t = await db.get(Transcript, transcript.id)
+            if not t:
+                _increment_files_completed(state)
+                return
+
+            t.status = "processing"
+            await db.commit()
+
+            # 计算该文件在总进度中的权重区间
+            # 预处理 0-5%, 判断+提取 5-30%, 切片提取 30-55%, 重组 55-60%, 生成 60-95%, 完成 100%
+            # 每个文件占 95/file_total 的权重
+            file_weight = 95.0 / max(file_total, 1)
+            file_start_percent = file_index * file_weight
+
+            def _file_percent(phase_start_ratio, phase_end_ratio, inner_ratio=1.0):
+                """计算当前文件在总进度中的百分比"""
+                phase_range = phase_end_ratio - phase_start_ratio
+                return min(99, int(file_start_percent + file_weight * (phase_start_ratio + phase_range * inner_ratio)))
+
+            state.update(
+                current_file=t.filename, action="正在读取文件...",
+                phase="preprocess", phase_percent=0, phase_detail="读取文件",
+            )
+
+            # 解析 DOCX
+            lines = await asyncio.to_thread(parse_transcript, t.file_path)
+            if not lines:
+                t.status = "completed"
+                t.error_message = "文件为空"
+                await db.commit()
+                _increment_files_completed(state)
+                return
+
+            # ========== 第一步：代码预处理（纯代码，不用 LLM）==========
+            state.update(
+                current_file=t.filename, action="正在分析对话结构...",
+                phase="preprocess", phase_percent=10, phase_detail="统计说话人",
+            )
+
+            preprocessor = Preprocessor()
+            filtered_lines = [p for p in lines if not preprocessor.is_management_line(p)]
+
+            if not filtered_lines:
+                t.status = "completed"
+                t.error_message = "过滤后无有效对话内容"
+                await db.commit()
+                _increment_files_completed(state)
+                return
+
+            segments, teacher = _preprocess_segments(filtered_lines)
+            logger.info(f"预处理完成: 识别老师={teacher}, 有效段落={len(segments)}")
+
+            if not segments:
+                t.status = "completed"
+                t.error_message = "未发现学员对话段落"
+                await db.commit()
+                _increment_files_completed(state)
+                return
+
+            state.update(
+                current_file=t.filename,
+                action=f"发现 {len(segments)} 个学员段落，开始识别故事...",
+                phase="preprocess", phase_percent=100, phase_detail="预处理完成",
+                percent=_file_percent(0, 0.05),
+            )
+
+            # ========== 第二步：判断故事 + 提取事实（并行）==========
+            state.update(
+                current_file=t.filename,
+                action=f"正在判断 {len(segments)} 个段落是否包含故事...",
+                phase="identify", phase_percent=0,
+                phase_detail=f"准备处理 {len(segments)} 个段落",
+            )
+
+            seg_semaphore = asyncio.Semaphore(MAX_CONCURRENT_SEGMENTS)
+            identify_results = [None] * len(segments)
+            identify_errors = []
+
+            async def process_segment_for_identify(seg_idx, speaker, seg):
+                """对单个段落进行判断+提取"""
+                async with seg_semaphore:
+                    text = _segment_to_text(seg)
+                    text_len = len(text)
+
+                    # 更新进度
+                    inner_pct = int((seg_idx + 1) / len(segments) * 100)
+                    state.update(
+                        phase="identify", phase_percent=inner_pct,
+                        phase_detail=f"正在处理段落 {seg_idx + 1}/{len(segments)}（{speaker}, {text_len}字）",
+                        percent=_file_percent(0.05, 0.30, (seg_idx + 1) / len(segments)),
+                    )
+
+                    if text_len <= SHORT_THRESHOLD:
+                        # 短段落：跳过判断，直接提取事实
+                        logger.info(f"  段落 {seg_idx}: {speaker} ({text_len}字) - 短段落，直接提取事实")
+                        facts = await _extract_facts_from_chunk(text, "")
+                        if facts:
+                            hints = facts.get("story_hints", "") or f"{speaker}的故事"
+                            return (speaker, seg, hints, [facts])
+                        else:
+                            logger.info(f"  段落 {seg_idx}: 提取失败，跳过")
+                            return None
+                    else:
+                        # 长段落：先判断是否有故事
+                        is_story, hints = await _identify_story(text)
+                        logger.info(f"  段落 {seg_idx}: {speaker} ({text_len}字) - is_story={is_story}, hints={hints}")
+                        if is_story:
+                            return (speaker, seg, hints, None)  # None 表示需要第三步切片提取
+                        else:
+                            logger.info(f"  段落 {seg_idx}: 无故事，跳过")
+                            return None
+
+            tasks = [
+                process_segment_for_identify(i, speaker, seg)
+                for i, (speaker, seg) in enumerate(segments)
+            ]
+            results = await asyncio.gather(*tasks, return_exceptions=True)
+
+            # 收集结果：分为两类
+            # - 直接提取到事实的（短段落）
+            # - 需要切片提取的（长段落，有故事）
+            all_facts = []  # 最终的 (speaker, seg, hints, facts_list) 列表
+            needs_chunking = []  # 需要进入第三步的 (speaker, seg, hints)
+
+            for i, r in enumerate(results):
+                if isinstance(r, Exception):
+                    error_str = str(r)
+                    logger.warning(f"段落 {i} 处理异常: {error_str}")
+                    identify_errors.append(error_str)
+                    if "529" in error_str or "overloaded" in error_str.lower() or "timed out" in error_str.lower():
+                        raise r
+                    continue
+                if r is None:
+                    continue
+                speaker, seg, hints, facts_list = r
+                if facts_list is not None:
+                    # 短段落已直接提取到事实
+                    all_facts.append((speaker, seg, hints, facts_list))
+                else:
+                    # 长段落有故事，需要切片提取
+                    needs_chunking.append((speaker, seg, hints))
+
+            if not all_facts and not needs_chunking:
+                t.status = "completed"
+                t.error_message = "未发现包含个人故事的段落"
+                await db.commit()
+                _increment_files_completed(state)
+                state.update(
+                    percent=_file_percent(0.05, 0.30, 1.0),
+                    action="未发现故事",
+                )
+                return
+
+            logger.info(f"第二步完成: 直接提取={len(all_facts)}, 需切片={len(needs_chunking)}")
+
+            # 第二步完成时已知道发现了多少个故事，立即更新
+            discovered = len(all_facts) + len(needs_chunking)
+            total_stories[0] += discovered
+            state.update(stories_found=total_stories[0])
+
+            # ========== 第三步：切片提取事实（并行）==========
+            if needs_chunking:
+                state.update(
+                    current_file=t.filename,
+                    action=f"正在切片提取 {len(needs_chunking)} 个故事...",
+                    phase="chunk", phase_percent=0,
+                    phase_detail=f"准备切片 {len(needs_chunking)} 个故事段落",
+                )
+
+                chunk_semaphore = asyncio.Semaphore(MAX_CONCURRENT_SEGMENTS)
+
+                async def process_chunked_story(story_idx, speaker, seg, hints):
+                    """对单个长段落进行切片提取"""
+                    async with chunk_semaphore:
+                        text = _segment_to_text(seg)
+                        chunks = _chunk_text_by_paragraphs(text, max_chars=CHUNK_MAX_CHARS, overlap_paragraphs=CHUNK_OVERLAP_PARAGRAPHS)
+                        logger.info(f"  故事 {story_idx}: {speaker} - 切成 {len(chunks)} 块")
+
+                        facts_list = []
+                        chunk_facts = await asyncio.gather(
+                            *[_extract_facts_from_chunk(chunk, hints) for chunk in chunks],
+                            return_exceptions=True,
+                        )
+                        for j, facts in enumerate(chunk_facts):
+                            inner_pct = int((story_idx * 10 + j + 1) / (len(needs_chunking) * 10) * 100)
+                            state.update(
+                                phase="chunk", phase_percent=inner_pct,
+                                phase_detail=f"正在切片提取 故事{story_idx + 1}/{len(needs_chunking)} 块{j + 1}/{len(chunks)}",
+                                percent=_file_percent(0.30, 0.55, (story_idx + j / max(len(chunks), 1)) / len(needs_chunking)),
+                            )
+                            if isinstance(facts, Exception):
+                                logger.warning(f"  块{j}提取异常: {facts}")
+                                continue
+                            if facts:
+                                facts_list.append(facts)
+
+                        return (speaker, seg, hints, facts_list)
+
+                chunk_tasks = [
+                    process_chunked_story(i, speaker, seg, hints)
+                    for i, (speaker, seg, hints) in enumerate(needs_chunking)
+                ]
+                chunk_results = await asyncio.gather(*chunk_tasks, return_exceptions=True)
+
+                for r in chunk_results:
+                    if isinstance(r, Exception):
+                        error_str = str(r)
+                        logger.warning(f"切片提取异常: {error_str}")
+                        if "529" in error_str or "overloaded" in error_str.lower() or "timed out" in error_str.lower():
+                            raise r
+                        continue
+                    if r is not None:
+                        all_facts.append(r)
+
+                logger.info(f"第三步完成: 总共 {len(all_facts)} 个故事段落完成事实提取")
+
+            state.update(
+                current_file=t.filename,
+                action="切片提取完成，正在重组素材...",
+                phase="chunk", phase_percent=100,
+                phase_detail="切片提取完成",
+                percent=_file_percent(0.30, 0.55, 1.0),
+            )
+
+            # ========== 第四步：重组（纯代码，无需 LLM）==========
+            state.update(
+                current_file=t.filename,
+                action="正在重组素材...",
+                phase="merge", phase_percent=50,
+                phase_detail="去重合并素材",
+                percent=_file_percent(0.55, 0.60, 0.5),
+            )
+
+            merged = _merge_facts(all_facts)
+            logger.info(f"第四步完成: 重组出 {len(merged)} 个故事素材")
+
+            if not merged:
+                t.status = "completed"
+                t.error_message = "素材重组后为空"
+                await db.commit()
+                _increment_files_completed(state)
+                return
+
+            state.update(
+                current_file=t.filename,
+                action=f"重组完成，正在生成 {len(merged)} 个故事...",
+                phase="merge", phase_percent=100,
+                phase_detail="重组完成",
+                percent=_file_percent(0.55, 0.60, 1.0),
+            )
+
+            # ========== 第五步：生成最终故事（并行）==========
+            state.update(
+                current_file=t.filename,
+                action=f"正在生成 {len(merged)} 个故事...",
+                phase="generate", phase_percent=0,
+                phase_detail=f"准备生成 {len(merged)} 个故事",
+                percent=_file_percent(0.60, 0.95, 0),
+            )
+
+            gen_semaphore = asyncio.Semaphore(MAX_CONCURRENT_SEGMENTS)
+
+            async def generate_one_story(story_idx, speaker, seg, material):
+                """生成单个故事并保存到数据库"""
+                async with gen_semaphore:
+                    inner_pct = int((story_idx + 1) / len(merged) * 100)
+                    state.update(
+                        phase="generate", phase_percent=inner_pct,
+                        phase_detail=f"正在生成故事 {story_idx + 1}/{len(merged)}（{speaker}）",
+                        percent=_file_percent(0.60, 0.95, (story_idx + 1) / len(merged)),
+                    )
+
+                    story_data = await _generate_story(material)
+                    if not story_data:
+                        logger.warning(f"  故事 {story_idx}: 生成失败")
+                        return None
+
+                    # 保存到数据库
+                    raw_content = story_data.get("content", {})
+                    if isinstance(raw_content, dict):
+                        content_str = _format_story_content(raw_content)
+                    else:
+                        content_str = str(raw_content)
+
+                    story = Story(
+                        title=story_data.get("title", "未命名故事"),
+                        summary=story_data.get("summary", ""),
+                        tags=json.dumps(story_data.get("tags", []), ensure_ascii=False),
+                        content=content_str,
+                        raw_material=json.dumps(material, ensure_ascii=False),
+                        speaker_nickname=story_data.get("speaker_nickname", ""),
+                        source_transcript_id=t.id,
+                        source_lines=json.dumps([l.get("index", 0) for l in seg if isinstance(l, dict)]),
+                        duration_minutes=_estimate_duration(seg),
+                        confidence=story_data.get("confidence", 0.7),
+                        confidence_level=_confidence_level(story_data.get("confidence", 0.7)),
+                    )
+                    db.add(story)
+                    await db.commit()
+
+                    state.update(
+                        action=f"生成第 {total_stories[0]} 个故事：{story.title}",
+                    )
+                    logger.info(f"  故事 {story_idx}: 保存成功 - {story.title}")
+
+                    return story
+
+            gen_tasks = [
+                generate_one_story(i, speaker, seg, material)
+                for i, (speaker, seg, material) in enumerate(merged)
+            ]
+            gen_results = await asyncio.gather(*gen_tasks, return_exceptions=True)
+
+            for r in gen_results:
+                if isinstance(r, Exception):
+                    error_str = str(r)
+                    logger.warning(f"故事生成异常: {error_str}")
+                    if "529" in error_str or "overloaded" in error_str.lower() or "timed out" in error_str.lower():
+                        raise r
+
+            t.status = "completed"
+            await db.commit()
+            _increment_files_completed(state)
+            state.update(
+                current_file=t.filename,
+                action=f"文件处理完成，共生成 {total_stories[0]} 个故事",
+                phase="generate", phase_percent=100,
+                phase_detail="文件处理完成",
+                percent=_file_percent(0.60, 0.95, 1.0),
+            )
+
+    except Exception as e:
+        error_str = str(e)
+        logger.exception(f"Error processing file {transcript.filename}")
+        try:
+            async with async_session() as db:
+                t = await db.get(Transcript, transcript.id)
+                if t:
+                    t.status = "error"
+                    t.error_message = error_str
+                    await db.commit()
+        except Exception:
+            pass
+        _increment_files_completed(state)
+
+        if "529" in error_str or "overloaded" in error_str.lower():
+            state.update(status="error", error="AI 服务暂时过载（529），请稍后再试")
+        elif "timed out" in error_str.lower():
+            state.update(status="error", error="AI 服务请求超时，请稍后再试")
+        else:
+            state.update(status="error", error=f"处理文件时出错：{error_str[:200]}")
+
+
+# ============================================================
+# 第一步：代码预处理
+# ============================================================
+def _preprocess_segments(lines: list[dict]) -> tuple[list[tuple[str, list[dict]]], str]:
+    """
+    代码预处理（v3：时间范围截取，允许重叠）。
+    识别老师 → 过滤杂音（<20句）→ 按时间范围截取 → 过滤短段落（<15句/<80字）→ 过滤非主讲述人（占比<10%）
+
+    返回: (segments, teacher)
+      segments: [(speaker, segment_lines), ...]
+      teacher: 老师名称
+    """
+    # 1. 统计每个说话人
+    speaker_stats = {}
+    for i, line in enumerate(lines):
+        speaker = line.get("speaker", "")
+        if not speaker:
+            continue
+        if speaker not in speaker_stats:
+            speaker_stats[speaker] = {"lines": 0, "chars": 0, "first": i, "last": i}
+        speaker_stats[speaker]["lines"] += 1
+        speaker_stats[speaker]["chars"] += len(line.get("content", ""))
+        speaker_stats[speaker]["last"] = i
+
+    # 2. 识别老师（发言最多的人）
+    teacher = max(speaker_stats, key=lambda s: speaker_stats[s]["lines"])
+    logger.info(f"识别老师: {teacher} ({speaker_stats[teacher]['lines']}行, {speaker_stats[teacher]['chars']}字)")
+
+    # 3. 过滤杂音（< 20句的说话人）+ 排除老师
+    students = []
+    noise_speakers = []
+    for s, stats in speaker_stats.items():
+        if s == teacher:
+            continue
+        if stats["lines"] < 20:
+            noise_speakers.append(f"{s} ({stats['lines']}句)")
+            continue
+        students.append(s)
+
+    if noise_speakers:
+        logger.info(f"杂音过滤: {', '.join(noise_speakers)}")
+    logger.info(f"有效学员: {students}")
+
+    # 4. 按时间范围截取：每个学员从第一句到最后一句，中间所有行全部截取
+    segments = []
+    for speaker in students:
+        stats = speaker_stats[speaker]
+        first_idx = stats["first"]
+        last_idx = stats["last"]
+        seg = lines[first_idx:last_idx + 1]
+        segments.append((speaker, seg))
+
+    # 5. 过滤：短段落 + 当事人占比过低（非主讲述人）
+    final_segments = []
+    for speaker, seg in segments:
+        student_lines = [l for l in seg if l.get("speaker", "") == speaker]
+        if len(student_lines) < 15:
+            logger.info(f"过滤: {speaker} ({len(student_lines)}句<15)")
+            continue
+        student_chars = sum(len(l.get("content", "")) for l in student_lines)
+        if student_chars < 80:
+            logger.info(f"过滤: {speaker} ({student_chars}字<80)")
+            continue
+        # 当事人说话占比 < 10% → 非主讲述人，丢弃
+        ratio = len(student_lines) / len(seg) if seg else 0
+        if ratio < 0.1:
+            logger.info(f"过滤: {speaker} (占比{ratio:.1%}<10%)")
+            continue
+        final_segments.append((speaker, seg))
+
+    logger.info(f"最终: {len(final_segments)} 个学员段落")
+    for i, (speaker, seg) in enumerate(final_segments):
+        student_lines = [l for l in seg if l.get("speaker", "") == speaker]
+        student_chars = sum(len(l.get("content", "")) for l in student_lines)
+        ratio = len(student_lines) / len(seg) if seg else 0
+        logger.info(f"  段落{i}: {speaker}, 截取{len(seg)}行, 学员{len(student_lines)}句/{student_chars}字, 占比{ratio:.1%}")
+
+    return final_segments, teacher
+
+
+# ============================================================
+# 第二步：判断故事
+# ============================================================
+async def _identify_story(text: str) -> tuple[bool, str]:
+    """判断对话段落是否包含个人故事"""
+    prompt = """判断这段对话是否包含学员的个人故事。直接输出JSON，不要分析。
+
+个人故事特征（满足至少2条）：学员讲述具体经历、包含时间地点人物细节、表达深层情感、老师深入引导疗愈、揭示心理或家庭问题。
+
+非故事内容：纯课堂讲解、简单问答寒暄、课堂管理、纯疗愈引导词。
+
+直接输出JSON：
+{"is_story": true或false, "confidence": 0.0到1.0, "story_hints": "一句话描述故事主题和讲述者"}
+
+对话内容：
+```
+""" + text[:4000] + """
+```"""
+
+    try:
+        result = await llm_client.chat_json([{"role": "user", "content": prompt}], temperature=0.2)
+        if result:
+            is_story = result.get("is_story", False)
+            confidence = result.get("confidence", 0)
+            hints = result.get("story_hints", "")
+            logger.info(f"Story identification: is_story={is_story}, confidence={confidence}, hints={hints[:80]}")
+            return bool(is_story) and confidence >= 0.5, hints
+    except Exception as e:
+        logger.warning(f"Story identification error: {e}")
+        if "529" in str(e) or "overloaded" in str(e).lower() or "timed out" in str(e).lower():
+            raise
+
+    return False, ""
+
+
+# ============================================================
+# 第三步：切片提取事实
+# ============================================================
+def _chunk_text_by_paragraphs(text: str, max_chars: int = 3000, overlap_paragraphs: int = 5) -> list[str]:
+    """按自然段切分，在 max_chars 附近找段落结尾切，重叠 overlap_paragraphs 个自然段"""
+    paragraphs = [p for p in text.split("\n") if p.strip()]
+    if not paragraphs:
+        return [text]
+    if len(text) <= max_chars:
+        return [text]
+
+    chunks = []
+    i = 0
+    while i < len(paragraphs):
+        # 从当前位置开始，累加段落直到超过 max_chars
+        current = []
+        current_len = 0
+        j = i
+        while j < len(paragraphs):
+            para = paragraphs[j]
+            # 超过 max_chars 且已有内容 → 在这个段落前切断（段落结尾切）
+            if current_len + len(para) > max_chars and current:
+                break
+            current.append(para)
+            current_len += len(para)
+            j += 1
+
+        chunks.append("\n".join(current))
+
+        # 到末尾了就结束
+        if j >= len(paragraphs):
+            break
+
+        # 下一次起点：当前切到的位置 - 重叠段落数
+        next_i = j - overlap_paragraphs
+        # 防止死循环：确保至少前进1个段落
+        if next_i <= i:
+            next_i = i + 1
+        i = next_i
+
+    return chunks
+
+
+async def _extract_facts_from_chunk(text: str, hints: str) -> dict | None:
+    """从对话切片中提取关键信息（事件、情感、原话、人物）"""
+    prompt = f"""从这段对话中提取关键信息。只输出JSON。
+
+故事主题：{hints}
+
+提取以下内容：
+- events: 发生的事件（尽量详细，保留具体细节）
+- emotions: 表达的情感
+- quotes: 讲述者的原话（保留原话，不要改写，尽量多提取；完全相同的重复语句只保留一条）
+- key_people: 提到的人物
+
+{{"events":["事件1","事件2"],"emotions":["情感1","情感2"],"quotes":["原话1","原话2","原话3"],"key_people":["人物1"]}}
+
+对话内容：
+```
+{text}
+```"""
+
+    try:
+        result = await llm_client.chat_json([{"role": "user", "content": prompt}], temperature=0.1)
+        if not result:
+            return None
+        # chat_json 可能返回 list（解析错误），需要转为 dict
+        if isinstance(result, list):
+            for item in result:
+                if isinstance(item, dict) and ("events" in item or "quotes" in item):
+                    result = item
+                    break
+            else:
+                logger.warning(f"chat_json returned list, cannot convert to dict: {result[:3]}")
+                return None
+        return result
+    except Exception as e:
+        logger.warning(f"Extract facts error: {e}")
+        if "529" in str(e) or "overloaded" in str(e).lower() or "timed out" in str(e).lower():
+            raise
+
+    return None
+
+
+# ============================================================
+# 第四步：重组（纯代码，无需 LLM）
+# ============================================================
+def _fuzzy_dedup_quotes(quotes: list[str]) -> list[str]:
+    """
+    模糊去重：对高度相似的语句只保留第一条。
+    策略：提取每条语句的核心部分（去掉常见前缀后缀），相似度超过阈值的视为重复。
+    """
+    import re
+
+    def normalize(s: str) -> str:
+        """提取语句核心，去掉语气词和细微差异"""
+        s = s.strip()
+        # 去掉「」包裹
+        if s.startswith("「") and s.endswith("」"):
+            s = s[1:-1]
+        # 去掉常见前缀
+        for prefix in ["嗯", "啊", "呃", "就是", "然后", "那个", "所以"]:
+            if s.startswith(prefix):
+                s = s[len(prefix):].strip()
+        # 去掉常见尾缀
+        for suffix in ["啊", "吧", "呢", "嘛", "哦", "呀"]:
+            if s.endswith(suffix) and len(s) > 4:
+                s = s[:-1]
+        # 去掉多余空白和重复字（如"我我我"、"的的的"）
+        s = re.sub(r'(.)\1{2,}', r'\1', s)
+        return s
+
+    def similarity(a: str, b: str) -> float:
+        """简单的字符级相似度（Jaccard on character bigrams）"""
+        if not a or not b:
+            return 0.0
+        a_bigrams = set(a[i:i+2] for i in range(len(a)-1))
+        b_bigrams = set(b[i:i+2] for i in range(len(b)-1))
+        if not a_bigrams or not b_bigrams:
+            return 0.0
+        intersection = a_bigrams & b_bigrams
+        union = a_bigrams | b_bigrams
+        return len(intersection) / len(union)
+
+    kept = []
+    kept_normalized = []
+    for q in quotes:
+        nq = normalize(q)
+        is_dup = False
+        for kn in kept_normalized:
+            # 短语句用高阈值，长语句用低阈值
+            threshold = 0.8 if len(nq) < 15 else 0.7
+            if similarity(nq, kn) >= threshold:
+                is_dup = True
+                break
+        if not is_dup:
+            kept.append(q)
+            kept_normalized.append(nq)
+
+    removed = len(quotes) - len(kept)
+    if removed > 0:
+        logger.info(f"    模糊去重: {len(quotes)} -> {len(kept)} (去除 {removed} 条相似重复)")
+
+    return kept
+
+
+def _merge_facts(all_facts: list) -> list[tuple[str, list, dict]]:
+    """
+    重组素材（Reduce）。
+    对每个故事段落的所有切片事实进行合并去重：
+    - 事件按原文顺序保留
+    - 情感去重
+    - 原话去重（不限制数量）
+    - 人物去重
+
+    输入: [(speaker, seg, hints, facts_list), ...]
+    输出: [(speaker, seg, material), ...]
+      material: {"events": [...], "emotions": [...], "quotes": [...], "key_people": [...], "story_hints": "..."}
+    """
+    merged = []
+    for i, (speaker, seg, hints, facts_list) in enumerate(all_facts):
+        all_events, all_emotions, all_quotes, all_people = [], [], [], []
+        for facts in facts_list:
+            if isinstance(facts, dict):
+                all_events.extend(facts.get("events", []))
+                all_emotions.extend(facts.get("emotions", []))
+                all_quotes.extend(facts.get("quotes", []))
+                all_people.extend(facts.get("key_people", []))
+
+        # 情感去重（保持顺序）
+        all_emotions = list(dict.fromkeys(all_emotions))
+        # 人物去重（保持顺序）
+        all_people = list(dict.fromkeys(all_people))
+        # 原话去重（保持顺序，不限制数量）
+        # 第一轮：精确去重
+        seen_quotes = set()
+        unique_quotes = [q for q in all_quotes if q not in seen_quotes and not seen_quotes.add(q)]
+        # 第二轮：模糊去重（去除重复模式的语句，如释放语句只保留一条）
+        unique_quotes = _fuzzy_dedup_quotes(unique_quotes)
+
+        material = {
+            "events": all_events,
+            "emotions": all_emotions,
+            "quotes": unique_quotes,
+            "key_people": all_people,
+            "story_hints": hints,
+        }
+
+        logger.info(f"  重组故事 {i}: {speaker} - 事件:{len(all_events)}, 情感:{len(all_emotions)}, "
+                     f"原话:{len(unique_quotes)}, 人物:{len(all_people)}")
+
+        merged.append((speaker, seg, material))
+
+    return merged
+
+
+# ============================================================
+# 第五步：生成最终故事
+# ============================================================
+async def _generate_story(material: dict) -> dict | None:
+    """基于素材生成第三人称故事（5部分结构化输出）"""
+    prompt = f"""基于素材生成第三人称故事。只输出JSON，不要其他内容。
+
+素材：
+主题：{material['story_hints']}
+事件：{json.dumps(material['events'], ensure_ascii=False)}
+情感：{json.dumps(material['emotions'], ensure_ascii=False)}
+原话：{json.dumps(material['quotes'], ensure_ascii=False)}
+人物：{json.dumps(material['key_people'], ensure_ascii=False)}
+
+写作手法（非常重要）：
+
+【双时空结构】
+故事有两个时空层：
+- "当下时空"：学员在课堂上与卢慧老师的对话——这是故事的框架，用对话体呈现，保留师生互动的完整脉络
+- "过往时空"：学员讲述的过去经历——这是故事的核心，用"场景重建"手法呈现，让读者穿越到那个时空
+
+【场景重建手法（用于"过往的故事"部分）】
+当学员讲述过去的经历时，不要写成"她说她小时候…"这种转述体，而是直接把读者拉进那个时空：
+- 直接切入场景，用过去时态叙述。比如不要写"文晔回忆说她的儿子15岁去了美国"，而要写"十五岁那年，文晔的儿子独自登上了飞往美国的航班"
+- 用五感描写构建画面：她看到了什么、听到了什么、身体感受到了什么（比如"手心出汗"、"脚发凉"这些身体反应要融入场景中）
+- 用环境细节暗示情绪：不要直接说"她很痛苦"，而是通过场景让读者感受到痛苦
+- 关键情感爆发处用直接引语（「」包裹原话），其他地方用间接引语和叙述自然融合
+- 场景之间要有过渡，比如"然而，命运的转折发生在前年冬天——"
+
+【对话体手法（用于"当下的困境"和"转变"部分）】
+- 保留师生对话的现场感，写出谁说了什么、对方如何回应
+- 在对话中穿插学员的内心活动、身体反应、情绪变化
+- 老师的引导语要完整保留，这是疗愈过程的关键
+
+【通用要求】
+- 每个事件都要展开写，写出具体过程和细节，不要一句话概括
+- 事件之间要有过渡和因果逻辑，不能跳跃
+- 情感变化要写出过程：从什么情感转变到什么情感，中间经历了什么
+- 不编造素材中没有的内容，但可以根据素材合理补充场景细节（如环境、氛围）
+- 每个部分充分展开，不要限制长度，越详细越好
+- 原文金句数组中的每条金句必须用英文双引号包裹，如 ["金句1", "金句2"]，不要用「」替代双引号
+
+禁止事项：
+- 禁止用一句话概括多个事件（如"她倾诉了挣扎"、"她经历了疗愈过程"）
+- 禁止省略素材中的具体细节
+- 禁止省略老师的引导和回应
+- 禁止事件之间跳跃，要保持时间顺序和因果逻辑
+- 禁止过度总结或抽象化描述
+- 禁止在"过往的故事"部分全程使用转述体（"她说…她回忆…"），必须用场景重建手法让读者身临其境
+- 禁止为了简洁而牺牲内容的丰富性
+
+输出格式：{{"title":"标题","summary":"80字摘要","tags":["标签"],"speaker_nickname":"昵称","content":{{"讲述者":"...","当下的困境":"...","过往的故事":"...","转变":"...","原文金句":["原话1","原话2","原话3"]}}}}"""
+
+    try:
+        result = await llm_client.chat_json(
+            [{"role": "user", "content": prompt}],
+            temperature=0.7,
+        )
+        if not result:
+            return None
+        # chat_json 可能返回 list（解析错误），需要转为 dict
+        if isinstance(result, list):
+            for item in result:
+                if isinstance(item, dict) and "title" in item:
+                    result = item
+                    break
+            else:
+                logger.warning(f"chat_json returned list, cannot convert to dict: {result[:3]}")
+                return None
+
+        # 解析 content 纯文本为结构化字段
+        raw_content = result.get("content", "")
+        if isinstance(raw_content, str) and "|||" in raw_content:
+            parts = raw_content.split("|||")
+            content_dict = {}
+            keys = ["讲述者", "当下的困境", "过往的故事", "转变", "原文金句"]
+            for idx, key in enumerate(keys):
+                content_dict[key] = parts[idx].strip() if idx < len(parts) else ""
+            result["content"] = content_dict
+        elif isinstance(raw_content, dict):
+            for key in ["讲述者", "当下的困境", "过往的故事", "转变", "原文金句"]:
+                if key not in raw_content:
+                    raw_content[key] = ""
+            result["content"] = raw_content
+        else:
+            result["content"] = {"完整故事": str(raw_content)}
+
+        if not result.get("tags"):
+            result["tags"] = []
+
+        return result
+    except Exception as e:
+        logger.warning(f"Generate story error: {e}")
+        if "529" in str(e) or "overloaded" in str(e).lower() or "timed out" in str(e).lower():
+            raise
+
+    return None
+
+
+# ============================================================
+# 工具函数
+# ============================================================
+def _segment_to_text(segment_lines: list) -> str:
+    """将段落行列表转为对话文本"""
+    parts = []
+    for line in segment_lines:
+        speaker = line.get("speaker", "")
+        content = line.get("content", "")
+        if speaker:
+            parts.append(f"{speaker}: {content}")
+        else:
+            parts.append(content)
+    return "\n".join(parts)
+
+
+def _increment_files_completed(state: TaskState):
+    """增加已完成文件计数"""
+    p = state.progress
+    files_completed = p.get("files_completed", 0) + 1
+    files_total = p.get("files_total", 1)
+    state.update(files_completed=files_completed)
+
+
+def _estimate_duration(segment_lines: list) -> float:
+    """根据时间戳估算故事时长（分钟）"""
+    timestamps = []
+    for line in segment_lines:
+        if isinstance(line, dict) and line.get("timestamp"):
+            try:
+                parts = line["timestamp"].split(":")
+                if len(parts) >= 2:
+                    minutes = int(parts[-2]) + int(parts[-1]) / 60
+                    if len(parts) == 3:
+                        minutes += int(parts[0]) * 60
+                    timestamps.append(minutes)
+            except (ValueError, IndexError):
+                continue
+
+    if len(timestamps) >= 2:
+        return round(max(timestamps) - min(timestamps), 1)
+    return 0
+
+
+def _confidence_level(confidence: float) -> str:
+    if confidence >= 0.8:
+        return "high"
+    elif confidence >= 0.6:
+        return "medium"
+    return "low"
+
+
+def _format_story_content(content_dict: dict) -> str:
+    """将 LLM 返回的结构化故事内容转为可读的 Markdown 格式"""
+    parts = []
+
+    # 按顺序输出各个章节
+    sections = ["讲述者", "当下的困境", "过往的故事", "转变"]
+    for section_name in sections:
+        if content_dict.get(section_name):
+            parts.append(f"## {section_name}\n\n{content_dict[section_name]}")
+
+    # 原文金句
+    quotes = content_dict.get("原文金句", [])
+    if not quotes:
+        quotes = content_dict.get("金句摘录", [])
+    if quotes and isinstance(quotes, list):
+        parts.append("## 原文金句")
+        for q in quotes:
+            parts.append(f"\n> {q}")
+
+    # 兼容旧格式：故事梗概
+    if content_dict.get("故事梗概") and "讲述者" not in content_dict:
+        parts.insert(0, f"## 故事梗概\n\n{content_dict['故事梗概']}")
+
+    # 兼容旧格式：关键要素
+    elements = content_dict.get("关键要素", [])
+    if elements and isinstance(elements, list) and "讲述者" not in content_dict:
+        parts.append("## 关键要素")
+        for elem in elements:
+            if isinstance(elem, dict):
+                title = elem.get("标题", "")
+                desc = elem.get("描述", "")
+                parts.append(f"\n### {title}\n\n{desc}")
+
+    # 如果上面都没匹配到，按通用 key-value 格式输出
+    if not parts:
+        for key, value in content_dict.items():
+            if isinstance(value, str):
+                parts.append(f"## {key}\n\n{value}")
+            elif isinstance(value, list):
+                parts.append(f"## {key}")
+                for item in value:
+                    if isinstance(item, dict):
+                        for k, v in item.items():
+                            parts.append(f"\n**{k}**：{v}")
+                    else:
+                        parts.append(f"\n- {item}")
+
+    return "\n\n".join(parts) if parts else json.dumps(content_dict, ensure_ascii=False)

backend/app/services/file_service.py

@@ -0,0 +1,63 @@
+import os
+import uuid
+import shutil
+from fastapi import UploadFile
+from app.config import settings
+
+
+class FileService:
+    """文件上传、存储、删除"""
+
+    def __init__(self):
+        os.makedirs(settings.UPLOAD_DIR, exist_ok=True)
+        os.makedirs(os.path.join(settings.UPLOAD_DIR, "transcripts"), exist_ok=True)
+        os.makedirs(os.path.join(settings.UPLOAD_DIR, "persons"), exist_ok=True)
+        os.makedirs(settings.EXPORT_DIR, exist_ok=True)
+
+    async def save_transcript(self, file: UploadFile) -> dict:
+        """保存上传的文字稿"""
+        file_id = str(uuid.uuid4())
+        ext = os.path.splitext(file.filename)[1]
+        save_name = f"{file_id}{ext}"
+        save_dir = os.path.join(settings.UPLOAD_DIR, "transcripts")
+        save_path = os.path.join(save_dir, save_name)
+
+        content = await file.read()
+        with open(save_path, "wb") as f:
+            f.write(content)
+
+        return {
+            "id": file_id,
+            "filename": file.filename,
+            "file_path": save_path,
+            "file_size": len(content),
+        }
+
+    async def save_person(self, file: UploadFile) -> dict:
+        """保存上传的讲述者信息"""
+        file_id = str(uuid.uuid4())
+        ext = os.path.splitext(file.filename)[1]
+        save_name = f"{file_id}{ext}"
+        save_dir = os.path.join(settings.UPLOAD_DIR, "persons")
+        save_path = os.path.join(save_dir, save_name)
+
+        content = await file.read()
+        with open(save_path, "wb") as f:
+            f.write(content)
+
+        return {
+            "id": file_id,
+            "filename": file.filename,
+            "file_path": save_path,
+            "file_size": len(content),
+        }
+
+    def delete_file(self, file_path: str) -> bool:
+        """删除文件"""
+        if os.path.exists(file_path):
+            os.remove(file_path)
+            return True
+        return False
+
+
+file_service = FileService()

backend/app/services/llm_client.py

@@ -0,0 +1,287 @@
+import json
+import logging
+import re
+import time
+
+from openai import AsyncOpenAI
+import anthropic
+
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+
+class LLMClient:
+    """LLM 客户端，支持 OpenAI 和 Anthropic SDK"""
+
+    def __init__(self):
+        self._openai_client = None
+        self._anthropic_client = None
+        self._api_key = settings.LLM_API_KEY
+        self._base_url = settings.LLM_BASE_URL
+        self._model = settings.LLM_MODEL
+        self._provider = settings.LLM_PROVIDER
+
+    def _get_openai_client(self) -> AsyncOpenAI:
+        if self._openai_client is None:
+            self._openai_client = AsyncOpenAI(
+                api_key=self._api_key,
+                base_url=self._base_url,
+                max_retries=0,
+                timeout=300.0,
+            )
+        return self._openai_client
+
+    def _get_anthropic_client(self) -> anthropic.AsyncAnthropic:
+        if self._anthropic_client is None:
+            # MiniMax Anthropic 端点
+            base_url = self._base_url.replace("/v1", "/anthropic").replace("api.minimax.chat", "api.minimaxi.com")
+            if not base_url.endswith("/anthropic"):
+                # 如果 base_url 不是标准格式，手动拼接
+                base_url = "https://api.minimaxi.com/anthropic"
+            self._anthropic_client = anthropic.AsyncAnthropic(
+                api_key=self._api_key,
+                base_url=base_url,
+                timeout=300.0,
+            )
+        return self._anthropic_client
+
+    def _use_anthropic(self) -> bool:
+        """判断是否使用 Anthropic SDK"""
+        return False  # MiniMax thinking 占用太多 token，统一用 OpenAI SDK
+
+    def update_config(self, base_url: str = None, api_key: str = None, model: str = None):
+        """更新配置（设置页面保存后调用）"""
+        if base_url:
+            self._base_url = base_url
+        if api_key:
+            self._api_key = api_key
+        if model:
+            self._model = model
+        # 重置客户端
+        self._openai_client = None
+        self._anthropic_client = None
+        # 更新 provider 判断
+        if base_url:
+            self._provider = "minimax" if "minimax" in base_url.lower() else "openai"
+
+    async def chat(self, messages: list[dict], temperature: float = None, max_tokens: int = 16384, timeout: float = 300.0) -> str:
+        """发送对话请求，返回文本内容（不含思考过程）"""
+        if temperature is None:
+            temperature = settings.TEMPERATURE
+
+        if self._use_anthropic():
+            return await self._chat_anthropic(messages, temperature, max_tokens, timeout)
+        else:
+            result = await self._chat_openai(messages, temperature, max_tokens, timeout)
+            return result["content"]
+
+    async def chat_detail(self, messages: list[dict], temperature: float = None, max_tokens: int = 16384, timeout: float = 300.0) -> dict:
+        """发送对话请求，返回详细信息（content + reasoning + finish_reason + usage + elapsed）"""
+        if temperature is None:
+            temperature = settings.TEMPERATURE
+
+        if self._use_anthropic():
+            # Anthropic 暂不支持 detail
+            content = await self._chat_anthropic(messages, temperature, max_tokens, timeout)
+            return {"content": content, "reasoning_content": "", "finish_reason": "stop", "usage": {}, "elapsed": 0}
+        else:
+            return await self._chat_openai(messages, temperature, max_tokens, timeout)
+
+    async def _chat_openai(self, messages: list[dict], temperature: float, max_tokens: int, timeout: float = 300.0) -> dict:
+        """OpenAI SDK 调用，返回详细信息 dict"""
+        client = self._get_openai_client()
+        t0 = time.time()
+        response = await client.chat.completions.create(
+            model=self._model,
+            messages=messages,
+            temperature=temperature,
+            max_tokens=max_tokens,
+            timeout=timeout,
+        )
+        elapsed = time.time() - t0
+
+        content = response.choices[0].message.content or ""
+        # 提取 reasoning_content（thinking）
+        reasoning = ""
+        msg = response.choices[0].message
+        if hasattr(msg, 'reasoning_content') and msg.reasoning_content:
+            reasoning = msg.reasoning_content
+        # 有些 SDK 把 reasoning 放在 model_extra 里
+        if not reasoning and hasattr(msg, 'model_extra') and isinstance(msg.model_extra, dict):
+            reasoning = msg.model_extra.get("reasoning_content", "")
+
+        # 检查是否有多个 choice
+        if len(response.choices) > 1:
+            print(f"[LLM] 警告: 返回了 {len(response.choices)} 个 choices")
+        # 检查 finish_reason
+        reason = response.choices[0].finish_reason if response.choices else "unknown"
+        if reason != "stop":
+            print(f"[LLM] 警告: finish_reason={reason}, 可能被截断")
+
+        # usage 信息
+        usage = {}
+        if hasattr(response, 'usage') and response.usage:
+            usage = {
+                "prompt_tokens": response.usage.prompt_tokens,
+                "completion_tokens": response.usage.completion_tokens,
+                "total_tokens": response.usage.total_tokens,
+            }
+            if hasattr(response.usage, 'completion_tokens_details') and response.usage.completion_tokens_details:
+                details = response.usage.completion_tokens_details
+                usage["reasoning_tokens"] = getattr(details, 'reasoning_tokens', 0) or 0
+
+        # 保存完整响应到文件，供调试
+        try:
+            resp_dict = response.model_dump() if hasattr(response, 'model_dump') else str(response)
+            debug_path = f"/tmp/llm_response_{id(response) % 100000}.json"
+            with open(debug_path, "w", encoding="utf-8") as _f:
+                json.dump(resp_dict, _f, ensure_ascii=False, indent=2, default=str)
+            print(f"[LLM] 完整响应已保存: {debug_path}")
+        except Exception:
+            pass
+
+        return {
+            "content": content,
+            "reasoning_content": reasoning,
+            "finish_reason": reason,
+            "usage": usage,
+            "elapsed": round(elapsed, 2),
+        }
+
+    async def _chat_anthropic(self, messages: list[dict], temperature: float, max_tokens: int, timeout: float = 300.0) -> str:
+        """Anthropic SDK 调用 - thinking 和 text 自动分离"""
+        client = self._get_anthropic_client()
+
+        # 转换消息格式：OpenAI -> Anthropic
+        system_msg = ""
+        anthropic_messages = []
+        for msg in messages:
+            role = msg.get("role", "user")
+            content = msg.get("content", "")
+            if role == "system":
+                system_msg = content
+            elif role in ("user", "assistant"):
+                anthropic_messages.append({"role": role, "content": content})
+
+        kwargs = {
+            "model": self._model,
+            "max_tokens": max_tokens,
+            "messages": anthropic_messages,
+            "thinking": {"type": "disabled"},  # 禁用 thinking，避免占用 token
+        }
+        if system_msg:
+            kwargs["system"] = system_msg
+        if temperature is not None:
+            kwargs["temperature"] = temperature
+
+        response = await client.messages.create(**kwargs)
+
+        # 从 response 中提取文本内容，thinking 自动分离在 block.thinking 中
+        text_parts = []
+        for block in response.content:
+            if hasattr(block, 'text') and block.type == 'text':
+                text_parts.append(block.text)
+            # thinking 内容自动忽略
+
+        result = "\n".join(text_parts)
+        return result
+
+    async def chat_json(self, messages: list[dict], temperature: float = None, max_tokens: int = 16384) -> dict:
+        """发送对话请求，期望返回 JSON"""
+        raw = await self.chat(messages, temperature, max_tokens=max_tokens)
+        return _parse_json(raw)
+
+    async def test_connection(self) -> dict:
+        """测试 LLM 连接"""
+        try:
+            response = await self.chat([{"role": "user", "content": "请回复：连接成功"}])
+            return {"success": True, "message": f"连接成功，模型回复：{response[:50]}"}
+        except Exception as e:
+            return {"success": False, "message": f"连接失败：{str(e)}"}
+
+
+def _parse_json(raw: str) -> dict:
+    """健壮的 JSON 解析，处理 LLM 返回的各种格式问题"""
+    print(f"[_parse_json] raw长度={len(raw)}, 前100字={raw[:100]}")
+    print(f"[_parse_json] raw后100字={raw[-100:]}")
+    raw = raw.strip()
+
+    # 去掉 markdown 代码块
+    if raw.startswith("```"):
+        raw = re.sub(r'^```\w*\n?', '', raw)
+        raw = re.sub(r'\n?```$', '', raw)
+        raw = raw.strip()
+
+    # 找 JSON 结构的位置
+    # 优先找第一个 { }（JSON 对象），用括号匹配确保完整
+    first_brace = raw.find('{')
+    last_bracket = raw.rfind('[')
+
+    start = -1
+    if first_brace >= 0:
+        start = first_brace
+        # 从第一个 { 开始，找匹配的 }
+        depth = 0
+        end = -1
+        for i in range(start, len(raw)):
+            if raw[i] == '{':
+                depth += 1
+            elif raw[i] == '}':
+                depth -= 1
+                if depth == 0:
+                    end = i + 1
+                    break
+        if end <= start:
+            start = -1
+    elif last_bracket >= 0:
+        start = last_bracket
+        end = raw.rfind(']') + 1
+        if end <= start:
+            start = -1
+
+    if start < 0:
+        start = raw.find('{')
+        end = raw.rfind('}') + 1
+
+    if start < 0 or end <= start:
+        raise ValueError(f"无法找到 JSON: {raw[:200]}")
+
+    json_str = raw[start:end]
+
+    # 修复常见问题：中文引号替换为英文引号（在JSON值内部的需要转义）
+    # 先把 JSON 结构中的英文引号保护起来，再替换中文引号
+    json_str = json_str.replace('\u201c', '\u201c').replace('\u201d', '\u201d')  # 先保持不变
+    json_str = json_str.replace('\u2018', "'").replace('\u2019', "'")
+    json_str = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f]', '', json_str)
+    json_str = re.sub(r',\s*([}\]])', r'\1', json_str)
+
+    try:
+        return json.loads(json_str)
+    except json.JSONDecodeError as e:
+        # 如果失败，尝试将中文引号替换为转义的英文引号
+        json_str2 = json_str.replace('\u201c', '\\"').replace('\u201d', '\\"')
+        try:
+            return json.loads(json_str2)
+        except json.JSONDecodeError:
+            pass
+
+    # 尝试把换行替换为 \n
+    json_str2 = json_str.replace('\n', '\\n').replace('\r', '\\r')
+    json_str2 = json_str2.replace('\\\\n', '\\n')
+    try:
+        return json.loads(json_str2)
+    except json.JSONDecodeError:
+        pass
+
+    # 尝试自动补全缺失的闭合括号（模型输出被截断时常见）
+    for extra in ['}', '}}', '}}}', '}]', '}]}']:
+        try:
+            return json.loads(json_str + extra)
+        except json.JSONDecodeError:
+            continue
+
+    raise ValueError(f"无法解析 JSON: {json_str[:300]}")
+
+
+llm_client = LLMClient()

backend/app/services/preprocessor.py

@@ -0,0 +1,220 @@
+import re
+
+
+class Preprocessor:
+    """
+    预处理器：从原始对话中提取包含故事的段落
+    
+    核心思路：
+    1. 过滤纯课堂管理内容（签到、纪律等）
+    2. 识别"个案对话段"——老师与某位学员的深度互动
+    3. 保留完整对话上下文（师生互动），不过度过滤
+    4. 按话题边界切分为合理的段落
+    """
+
+    # 课堂管理关键词（出现这些词的整行直接过滤）
+    MANAGEMENT_KEYWORDS = ["签到", "打卡", "禁言", "扣三个一", "进组", "会议室链接"]
+
+    # 纯课堂管理发言模式（整条消息都是管理内容）
+    MANAGEMENT_PATTERNS = [
+        re.compile(r"大家.*签到"),
+        re.compile(r"请.*进组"),
+        re.compile(r"会议室"),
+        re.compile(r"纪律"),
+        re.compile(r"昵称.*改为"),
+        re.compile(r"欢迎.*进入"),
+    ]
+
+    # 老师关键词
+    TEACHER_KEYWORDS = ["卢慧老师", "静静", "督导"]
+
+    def is_teacher(self, speaker: str) -> bool:
+        """判断是否为老师/助教"""
+        return any(kw in speaker for kw in self.TEACHER_KEYWORDS)
+
+    def is_management_line(self, line: dict) -> bool:
+        """判断是否为课堂管理行"""
+        content = line.get("content", "")
+        # 短内容 + 管理关键词 = 管理行
+        if len(content) < 30:
+            return any(kw in content for kw in self.MANAGEMENT_KEYWORDS)
+        return any(p.search(content) for p in self.MANAGEMENT_PATTERNS)
+
+    def preprocess(self, paragraphs: list[dict]) -> list[dict]:
+        """
+        完整预处理流程：
+        1. 过滤课堂管理
+        2. 识别个案对话段（保留师生互动）
+        3. 按主要讲述者拆分（每个段落只包含一个主要讲述者）
+        """
+        # 第一步：过滤课堂管理行
+        filtered = [p for p in paragraphs if not self.is_management_line(p)]
+        
+        # 第二步：识别个案对话段
+        segments = self._extract_story_segments(filtered)
+        
+        # 第三步：按主要讲述者拆分
+        segments = self._split_by_speaker(segments)
+        
+        return segments
+
+    def _split_by_speaker(self, segments: list[list[dict]]) -> list[list[dict]]:
+        """
+        将每个段落按讲述者拆分成子段落。
+        同一个讲述者的所有对话（即使中间隔了其他人的插话）收集在一起，
+        老师发言分配给前一个活跃的学员作为上下文。
+        """
+        result = []
+        for seg in segments:
+            # 按讲述者收集对话，保留老师发言作为上下文
+            speaker_blocks = {}  # speaker -> [lines]
+            last_student = None
+
+            for line in seg:
+                speaker = line.get("speaker", "")
+                if self.is_teacher(speaker):
+                    # 老师发言归入前一个活跃的学员
+                    if last_student and last_student in speaker_blocks:
+                        speaker_blocks[last_student].append(line)
+                else:
+                    if speaker not in speaker_blocks:
+                        speaker_blocks[speaker] = []
+                    speaker_blocks[speaker].append(line)
+                    last_student = speaker
+
+            # 每个讲述者的对话作为一个子段落
+            for speaker, lines in speaker_blocks.items():
+                student_content = sum(
+                    len(l.get("content", ""))
+                    for l in lines
+                    if not self.is_teacher(l.get("speaker", ""))
+                )
+                # 过滤掉内容太少的（可能是插话而非讲述者）
+                if student_content > 80 and len(lines) >= 3:
+                    # 计算该讲述者在自己段落中的字数占比
+                    # 如果老师发言占比超过50%，说明这主要是老师在讲课，不是学员故事
+                    teacher_chars = sum(
+                        len(l.get("content", ""))
+                        for l in lines
+                        if self.is_teacher(l.get("speaker", ""))
+                    )
+                    total_chars = student_content + teacher_chars
+                    if total_chars > 0 and teacher_chars / total_chars > 0.5:
+                        continue  # 老师讲太多，跳过
+                    result.append(lines)
+
+        return result
+
+    def _extract_story_segments(self, paragraphs: list[dict]) -> list[list[dict]]:
+        """
+        从过滤后的对话中提取可能包含故事的段落。
+        
+        策略：
+        - 找到学员的"长发言"（>50字）作为故事线索
+        - 向前向后扩展，包含老师的引导和互动
+        - 直到遇到另一位学员的长发言或长时间沉默
+        - 最小段落 10 行，最大段落 200 行
+        """
+        segments = []
+        i = 0
+        used = set()  # 已分配的行索引
+
+        while i < len(paragraphs):
+            p = paragraphs[i]
+            
+            # 跳过老师单独发言（非互动）
+            if self.is_teacher(p.get("speaker", "")):
+                i += 1
+                continue
+            
+            # 跳过短发言（< 30字不太可能开启故事）
+            if len(p.get("content", "")) < 30:
+                i += 1
+                continue
+            
+            # 跳过已分配的行
+            if p.get("index", i) in used:
+                i += 1
+                continue
+
+            # 找到一个学员发言，开始扩展
+            segment_start = i
+            segment_end = i
+            
+            # 向前扩展（找老师之前的引导）
+            expand_start = max(0, i - 5)
+            for j in range(i - 1, expand_start - 1, -1):
+                if j in used:
+                    break
+                prev = paragraphs[j]
+                # 如果前一个是老师发言，包含
+                if self.is_teacher(prev.get("speaker", "")):
+                    segment_start = j
+                # 如果前一个是同一学员的发言，包含
+                elif prev.get("speaker") == p.get("speaker"):
+                    segment_start = j
+                else:
+                    # 其他学员发言，停止
+                    break
+
+            # 向后扩展（找老师的回应和后续互动）
+            current_speaker = p.get("speaker", "")
+            for j in range(i + 1, min(len(paragraphs), i + 200)):
+                if j in used:
+                    segment_end = j - 1
+                    break
+                    
+                curr = paragraphs[j]
+                curr_speaker = curr.get("speaker", "")
+                curr_content = curr.get("content", "")
+                
+                # 老师发言：包含（互动引导）
+                if self.is_teacher(curr_speaker):
+                    segment_end = j
+                    continue
+                
+                # 同一学员继续发言：包含
+                if curr_speaker == current_speaker:
+                    segment_end = j
+                    continue
+                
+                # 其他学员短回应（< 20字）：可能是插话，跳过但不终止
+                if len(curr_content) < 20:
+                    continue
+                
+                # 其他学员长发言：新的话题开始，终止
+                break
+
+            # 提取段落
+            segment = paragraphs[segment_start:segment_end + 1]
+            
+            # 过滤掉太短的段落（< 8行）
+            if len(segment) >= 8:
+                # 检查是否有足够的学员实质性内容（非纯疗愈练习）
+                student_content_len = sum(
+                    len(s.get("content", "")) 
+                    for s in segment 
+                    if not self.is_teacher(s.get("speaker", ""))
+                    and len(s.get("content", "")) > 15
+                )
+                
+                # 学员实质内容 > 100字才认为是故事
+                if student_content_len > 100:
+                    segments.append(segment)
+                    for s in segment:
+                        used.add(s.get("index", 0))
+                    i = segment_end + 1
+                    continue
+            
+            i += 1
+
+        return segments
+
+
+def is_teacher(speaker: str) -> bool:
+    """判断是否为老师/助教发言"""
+    teacher_keywords = ["卢慧老师", "静静", "督导", "老师"]
+    return any(kw in speaker for kw in teacher_keywords)
+
+
+preprocessor = Preprocessor()

backend/app/state.py

@@ -0,0 +1,95 @@
+import asyncio
+import json
+import uuid
+from dataclasses import dataclass, field
+from typing import Optional
+
+
+@dataclass
+class TaskState:
+    """单个提取任务的状态"""
+
+    task_id: str = ""
+    is_running: bool = False
+    cancel_event: asyncio.Event = field(default_factory=asyncio.Event)
+    progress: dict = field(default_factory=lambda: {
+        "percent": 0,
+        "status": "idle",
+        "current_file": "",
+        "action": "",
+        "files_total": 0,
+        "files_completed": 0,
+        "stories_found": 0,
+        "phase": "",
+        "phase_percent": 0,
+        "phase_detail": "",
+    })
+    _queue: asyncio.Queue = field(default_factory=asyncio.Queue)
+
+    def update(self, **kwargs):
+        """更新进度，同时推送到 SSE 队列"""
+        self.progress.update(kwargs)
+        try:
+            self._queue.put_nowait(self.progress.copy())
+        except asyncio.QueueFull:
+            pass
+
+    async def get_update(self) -> dict:
+        """SSE 消费端：等待新的进度更新"""
+        return await self._queue.get()
+
+    def reset(self):
+        """重置状态"""
+        self.is_running = False
+        self.cancel_event = asyncio.Event()
+        self.progress = {
+            "percent": 0,
+            "status": "idle",
+            "current_file": "",
+            "action": "",
+            "files_total": 0,
+            "files_completed": 0,
+            "stories_found": 0,
+            "phase": "",
+            "phase_percent": 0,
+            "phase_detail": "",
+        }
+        self._queue = asyncio.Queue()
+
+
+class TaskManager:
+    """多任务管理器：支持多个用户同时提取"""
+
+    def __init__(self):
+        self._tasks: dict[str, TaskState] = {}
+
+    def create_task(self) -> TaskState:
+        """创建新任务"""
+        task_id = uuid.uuid4().hex[:12]
+        state = TaskState(task_id=task_id)
+        self._tasks[task_id] = state
+        return state
+
+    def get_task(self, task_id: str) -> Optional[TaskState]:
+        """获取任务状态"""
+        return self._tasks.get(task_id)
+
+    def remove_task(self, task_id: str):
+        """清理已完成的任务"""
+        self._tasks.pop(task_id, None)
+
+    def cleanup_old_tasks(self):
+        """清理所有非运行中的任务"""
+        to_remove = [
+            tid for tid, state in self._tasks.items()
+            if not state.is_running
+        ]
+        for tid in to_remove:
+            self._tasks.pop(tid, None)
+
+
+# 全局任务管理器
+task_manager = TaskManager()
+
+# 向后兼容：默认任务（单用户场景）
+extraction_state = TaskState()

backend/requirements.txt

@@ -0,0 +1,9 @@
+fastapi>=0.110.0
+uvicorn>=0.27.0
+python-multipart>=0.0.9
+sqlalchemy>=2.0
+aiosqlite>=0.19.0
+python-docx>=1.1.0
+openai>=1.12.0
+pydantic>=2.0
+python-dotenv>=1.0.0

build.py

@@ -0,0 +1,144 @@
+"""
+故事提取工具 - 打包构建脚本
+使用方法: python build.py
+"""
+import os
+import sys
+import shutil
+import subprocess
+
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+BACKEND_DIR = os.path.join(BASE_DIR, "backend")
+FRONTEND_DIR = os.path.join(BASE_DIR, "frontend")
+DIST_DIR = os.path.join(BASE_DIR, "dist")
+BUILD_DIR = os.path.join(BASE_DIR, "build")
+
+
+def step(msg: str):
+    print(f"\n{'='*50}")
+    print(f"  {msg}")
+    print(f"{'='*50}")
+
+
+def build_frontend():
+    """构建前端"""
+    step("1/4 构建前端...")
+    os.chdir(FRONTEND_DIR)
+    subprocess.run([sys.executable, "-m", "npm", "run", "build"], check=True, shell=False)
+    os.chdir(BASE_DIR)
+    print("前端构建完成！")
+
+
+def create_package_structure():
+    """创建打包目录结构"""
+    step("2/4 准备打包文件...")
+
+    # 清理旧的 dist
+    if os.path.exists(DIST_DIR):
+        shutil.rmtree(DIST_DIR)
+    os.makedirs(DIST_DIR, exist_ok=True)
+
+    # 复制 backend 目录
+    pkg_backend = os.path.join(DIST_DIR, "backend")
+    if os.path.exists(pkg_backend):
+        shutil.rmtree(pkg_backend)
+    shutil.copytree(BACKEND_DIR, pkg_backend)
+
+    # 复制 frontend/dist
+    pkg_frontend = os.path.join(DIST_DIR, "frontend", "dist")
+    os.makedirs(pkg_frontend, exist_ok=True)
+    src_dist = os.path.join(FRONTEND_DIR, "dist")
+    if os.path.exists(src_dist):
+        for item in os.listdir(src_dist):
+            s = os.path.join(src_dist, item)
+            d = os.path.join(pkg_frontend, item)
+            if os.path.isdir(s):
+                shutil.copytree(s, d)
+            else:
+                shutil.copy2(s, d)
+
+    # 复制 run.py
+    shutil.copy2(os.path.join(BASE_DIR, "run.py"), os.path.join(DIST_DIR, "run.py"))
+
+    print("打包文件准备完成！")
+
+
+def build_exe():
+    """使用 PyInstaller 打包"""
+    step("3/4 打包成 exe...")
+
+    os.chdir(DIST_DIR)
+
+    cmd = [
+        sys.executable, "-m", "PyInstaller",
+        "--name=故事提取工具",
+        "--onefile",
+        "--windowed",
+        "--noconfirm",
+        "--clean",
+        # 隐藏导入
+        "--hidden-import=uvicorn.logging",
+        "--hidden-import=uvicorn.loops",
+        "--hidden-import=uvicorn.loops.auto",
+        "--hidden-import=uvicorn.protocols",
+        "--hidden-import=uvicorn.protocols.http",
+        "--hidden-import=uvicorn.protocols.http.auto",
+        "--hidden-import=uvicorn.protocols.websockets",
+        "--hidden-import=uvicorn.protocols.websockets.auto",
+        "--hidden-import=uvicorn.lifespan",
+        "--hidden-import=uvicorn.lifespan.on",
+        "--hidden-import=multipart",
+        # 收集数据文件
+        f"--add-data=backend;backend",
+        f"--add-data=frontend;frontend",
+        # 入口
+        "run.py",
+    ]
+
+    subprocess.run(cmd, check=True, shell=False)
+    os.chdir(BASE_DIR)
+    print("打包完成！")
+
+
+def copy_output():
+    """复制最终输出"""
+    step("4/4 整理输出...")
+
+    exe_name = "故事提取工具.exe"
+    src = os.path.join(DIST_DIR, "dist", exe_name)
+    dst = os.path.join(BASE_DIR, exe_name)
+
+    if os.path.exists(src):
+        shutil.copy2(src, dst)
+        size_mb = os.path.getsize(dst) / (1024 * 1024)
+        print(f"\n✅ 打包成功！")
+        print(f"   输出文件: {dst}")
+        print(f"   文件大小: {size_mb:.1f} MB")
+        print(f"\n   使用方法:")
+        print(f"   1. 将 {exe_name} 发送给用户")
+        print(f"   2. 用户双击运行，自动打开浏览器")
+        print(f"   3. 首次使用需要在设置页面配置 API Key")
+    else:
+        print(f"\n❌ 打包失败，未找到输出文件: {src}")
+
+
+def main():
+    print("故事提取工具 - 打包构建")
+    print("=" * 50)
+
+    # 检查 PyInstaller
+    try:
+        import PyInstaller
+    except ImportError:
+        print("正在安装 PyInstaller...")
+        subprocess.run([sys.executable, "-m", "pip", "install", "pyinstaller", "--break-system-packages"],
+                       check=True, shell=False)
+
+    build_frontend()
+    create_package_structure()
+    build_exe()
+    copy_output()
+
+
+if __name__ == "__main__":
+    main()

docker-compose.yml

@@ -0,0 +1,14 @@
+version: '3.8'
+
+services:
+  app:
+    build: .
+    ports:
+      - "80:8000"
+    volumes:
+      - ./data/uploads:/app/uploads
+      - ./data/exports:/app/exports
+      - ./data/db:/app/data
+    env_file:
+      - backend/.env
+    restart: unless-stopped

docs/Story-Extractor-产品设计文档.md

@@ -0,0 +1,410 @@
+# Story Extractor — 产品设计文档
+
+> 版本：v1.1 · 2026-04-21
+
+---
+
+## 一、产品概述
+
+### 1.1 产品定位
+
+Story Extractor 是一个 **局域网部署的 Web 工具**，用于从直播回放文字稿中自动提取学员个人故事，与讲述者个人信息匹配后，为每位讲述者生成独立的 DOCX 文档。
+
+### 1.2 核心价值
+
+- **自动化**：AI 自动从上万行口语化文字稿中提取学员故事，替代人工阅读筛选
+- **结构化**：将散乱的直播内容转化为结构化的"人物 + 故事"文档
+- **可复用**：工具化交付，支持后续扩展（LLM 升级、搜索引擎等）
+
+### 1.3 使用场景
+
+| 维度 | 描述 |
+|------|------|
+| **部署环境** | 局域网，多人可访问 |
+| **数据规模** | 20+ 份文字稿（每份上万行），8+ 位讲述者 |
+| **用户角色** | 内容运营人员 |
+
+---
+
+## 二、文字稿特征分析
+
+### 2.1 格式特征
+
+每段格式为：**`发言者(时间戳): 内容`**，例如：
+```
+卢慧老师(00:05:38): 这个我们在我们的启蒙课里面...
+14班+HCM(02:39:31): 就是要要打要骂，呵呵...
+```
+
+- 有明确的**说话人标记**（老师 / 学员昵称）
+- 有**时间戳**，可用于判断发言时长
+- 主讲老师占 60-70% 内容，学员发言散落其中
+
+### 2.2 内容构成
+
+| 内容类型 | 占比 | 说明 |
+|----------|------|------|
+| **课程讲授** | ~55% | 老师讲理论、概念、方法论 |
+| **课堂管理** | ~10% | 改昵称、纪律提醒、进组讨论等技术性内容 |
+| **疗愈练习** | ~15% | "我愿意看见且释放..." 引导式重复 |
+| **学员故事** | ~15% | 学员讲述个人经历，老师追问 |
+| **学员短回应** | ~5% | "好的""收到"等 |
+
+### 2.3 故事的实际形态
+
+故事**不是一段完整的自述**，而是**对话式展开**的：
+- 老师提问/引导 → 学员回答 → 老师追问 → 学员深入讲述 → 可能穿插疗愈练习
+- 一个故事可能跨 **20-30 段对话**，持续 15-30 分钟
+- 故事之间没有明显分界，可能上一段还在讲理论，下一段就切入学员故事
+
+### 2.4 典型故事示例
+
+| 讲述者 | 故事概要 |
+|--------|----------|
+| **14班+HCM** | 从小被妈妈体罚（跪凳子），长大后用同样方式对待女儿，女儿出现躯体化症状 |
+| **13-阿纳** | 姥姥因特殊年代藏首饰恐惧自杀，形成"不能有钱"的家族限制性信念 |
+| **06班Rena** | 妈妈在怀自己前打掉一个男孩，爸爸重男轻女，自己被当男孩养，对亲密关系有破坏倾向 |
+| **11班四维** | 女儿出柜说喜欢同性，追溯到自己出生时妈妈因生女孩而悲伤的家族重男轻女 |
+
+---
+
+## 三、用户流程
+
+### 3.1 流程总览
+
+```
+欢迎页 → 上传文件 → 提取故事 → 匹配确认 → 预览导出
+```
+
+采用 **线性向导式** 交互，用户依次完成四个步骤。
+
+### 3.2 步骤指示器
+
+- 顶部居中展示 4 个步骤节点（上传文件 → 提取故事 → 匹配确认 → 预览导出）
+- 当前步骤高亮（紫色），已完成步骤显示绿色 ✓
+- 步骤之间用连线表示进度
+- 欢迎页不显示步骤指示器
+- 步骤节点可点击跳转（已完成的步骤）
+
+### 3.3 底部导航栏
+
+- 固定在页面底部
+- 左侧：上一步按钮（第一步时隐藏）
+- 中间：当前步骤提示文案
+- 右侧：下一步按钮（最后一步变为"全部导出"）
+- 欢迎页不显示底部导航栏
+
+---
+
+## 四、页面详细设计
+
+### 4.0 欢迎页
+
+**目的**：产品介绍 + 引导用户开始
+
+**布局**：页面垂直居中
+
+**内容**：
+- 产品 Logo（紫色渐变圆角方块 + 书本图标）
+- 产品名称 "Story Extractor"
+- 一句话介绍："从直播回放文字稿中，AI 自动提取个人故事，与讲述者信息匹配，一键生成精美文档。"
+- 四步流程预览（图标 + 标题 + 副标题）：
+  - 📤 上传文件 — 文字稿 + 个人信息
+  - 🪄 提取故事 — AI 智能识别
+  - 🔗 匹配确认 — 人工核对
+  - 📄 预览导出 — 生成 DOCX
+- 大号 CTA 按钮："开始使用"
+
+**交互**：
+- 点击"开始使用"进入第一步（上传文件）
+- 不显示步骤指示器和底部导航栏
+
+---
+
+### 4.1 第一步：上传文件
+
+**目的**：上传直播文字稿和讲述者个人信息
+
+**布局**：左右两栏
+
+**左栏 — 直播文字稿**：
+- 标题：直播文字稿（蓝色图标）+ 右侧标注 "DOCX 格式"
+- 上传区域（虚线边框拖拽区）：
+  - 图标 + "拖拽文件到此处，或点击上传"
+  - 提示：支持批量上传 · 仅限 .docx 格式
+- 已上传文件列表：
+  - 每个文件显示：文件图标（蓝色 DOCX）+ 文件名 + 元信息（行数、大小）
+  - 右侧删除按钮
+
+**右栏 — 讲述者信息**：
+- 标题：讲述者信息（粉色图标）+ 右侧标注 "一人一个 DOCX"
+- 上传区域（虚线边框拖拽区）：
+  - 图标 + "拖拽文件到此处，或点击上传"
+  - 提示：每个讲述者一个 DOCX · 包含个人信息和照片
+- 已上传文件列表：
+  - 每个文件显示：文件图标（粉色）+ 文件名 + 元信息（含几张照片）
+  - 右侧预览按钮 + 删除按钮
+
+**交互**：
+- 底部导航提示："上传文字稿和讲述者信息文件后进入下一步"
+- 点击"下一步"进入第二步
+
+---
+
+### 4.2 第二步：提取故事
+
+**目的**：AI 自动从所有文字稿中提取学员个人故事
+
+**进入方式**：从第一步点击"下一步"后自动进入，**自动开始提取**，无需手动触发
+
+#### 阶段 A：提取进度（自动展示）
+
+**布局**：单卡片
+
+**内容**：
+- 标题："AI 正在工作中" + 右侧百分比（大号加粗）
+- 进度条（粗圆角，紫色渐变填充）
+- 当前处理状态：
+  - 运行状态指示灯（紫色脉冲动画）
+  - 当前文件名："正在读取：xxx.docx"
+  - 当前动作：
+    - "预处理：过滤课堂管理内容..."
+    - "预处理：过滤疗愈练习..."
+    - "智能分段：识别话题边界..."
+    - "故事识别：检测学员发言..."
+    - "故事提取：整理故事内容..."
+- 三列状态卡片：
+  - 文件读取：`X / 23`（已读文件数 / 总文件数）
+  - 话题分段：`XX 段`（已分段数量）
+  - 故事发现：`X 个`（已发现故事数量，发现后变绿色）
+
+**进度阶段**：
+1. 0-15%：读取文件，预处理（过滤课堂管理、短回应）
+2. 15-40%：智能话题分段
+3. 40-70%：识别学员故事片段
+4. 70-95%：提取故事内容，生成标题和摘要
+5. 95-100%：汇总结果
+
+**完成时**：
+- 进度条变绿色
+- 状态灯变绿色
+- 显示"全部处理完成！共处理 X 份文字稿，发现 X 个故事"
+- 1.2 秒后自动切换到故事列表
+
+#### 阶段 B：故事列表（提取完成后展示）
+
+**布局**：单卡片
+
+**内容**：
+- 标题："已提取的故事" + 右侧 "共 X 个故事"
+- 故事卡片列表，每个卡片包含：
+  - 故事标题（加粗）
+  - 讲述者昵称（从文字稿提取，如"14班+HCM"）
+  - 故事摘要（2-3 句话）
+  - 来源信息（文件名 + 行号范围 + 时长）
+  - 置信度标签（高/中/低）
+  - 操作按钮：查看原文、编辑、删除
+
+**交互**：
+- 底部导航提示："AI 已提取故事，下一步将故事与讲述者匹配"
+- 点击"下一步"进入第三步（**无需全部匹配，可直接进入**）
+
+---
+
+### 4.3 第三步：匹配确认
+
+**目的**：将提取的故事与讲述者一一对应（可选操作）
+
+**布局**：三栏（故事列表 | 匹配操作 | 讲述者列表）
+
+**左栏 — 故事列表**：
+- 标题："故事列表" + 右侧 "X 个故事"
+- Tab 切换：全部 / 已匹配 / 未匹配
+- 故事卡片列表：
+  - 故事标题
+  - 讲述者昵称（文字稿中的昵称，如"14班+HCM"）
+  - 故事摘要（简短版）
+  - 匹配状态标签（已匹配显示讲述者姓名，未匹配显示"待匹配"）
+  - 点击选中（高亮为橙色背景）
+  - 操作：删除（不满意的故事可直接删除）
+
+**中间 — 匹配操作**：
+- 紫色圆形箭头按钮
+- 提示文字："选中故事和讲述者后点击匹配"
+
+**右栏 — 讲述者列表**：
+- 标题："讲述者" + 右侧 "共 X 人"
+- 讲述者卡片列表：
+  - 头像（姓氏首字，紫色渐变背景）
+  - 姓名
+  - 文字稿昵称映射（如：张小花 → 14班+HCM）
+  - 匹配状态（已匹配 X 个故事 / 未匹配）
+  - 点击选中（紫色边框高亮）
+
+**交互**：
+- 选中左侧故事 + 选中右侧讲述者 → 点击中间箭头确认匹配
+- 匹配完成后故事状态变为"已匹配"
+- **可删除不满意的故事**（AI 可能提取出无效故事）
+- **无需全部匹配即可进入下一步**（未匹配的故事不会导出）
+- 底部导航提示："将需要的故事与讲述者匹配，未匹配的故事不会导出"
+
+---
+
+### 4.4 第四步：预览导出
+
+**目的**：预览最终文档效果并导出
+
+**布局**：左右两栏
+
+**左栏 — 导出列表**：
+- 标题："导出列表" + 右侧 "仅显示已匹配的讲述者"
+- 导出项列表，每项包含：
+  - 状态图标（绿色 ✓ = 已匹配可导出）
+  - 文件名："张小花 - 故事集.docx"
+  - 元信息："X 个故事 · 含个人信息和照片"
+  - 预览按钮、下载按钮
+- 未匹配的故事不会出现在导出列表中
+
+**右栏 — 文档预览**：
+- 标题："文档预览" + 上一个/下一个切换按钮
+- 文档预览卡片（模拟 DOCX 效果）：
+  - 顶部：紫色渐变头部
+    - 圆形头像占位
+    - 讲述者姓名
+    - 个人信息（性别 · 年龄 · 城市 · 职业）
+  - 正文区域：
+    - "她的故事" 标题（紫色下划线）
+    - 故事正文（多段落，如有多个故事依次排列）
+
+**交互**：
+- 点击"上一个/下一个"切换预览不同讲述者的文档
+- 点击单个下载按钮导出单个 DOCX
+- 底部"全部导出"按钮（绿色）批量导出所有已匹配讲述者的文档
+
+---
+
+### 4.5 系统设置（弹窗）
+
+**触发**：右上角齿轮图标
+
+**布局**：居中弹窗 Modal
+
+**内容**：
+
+**LLM 配置**：
+- API 提供商（下拉选择）：OpenAI (GPT-4) / DeepSeek / 通义千问 / 本地 Ollama / 自定义兼容 API
+- 模型名称（文本输入）
+- API Base URL（文本输入）
+- API Key（密码输入）
+
+**提取参数**：
+- 话题分段最大长度（行数）
+- 故事最小长度（行数）
+- 置信度阈值（低于此值的故事标记为"低置信度"）
+- 并发请求数
+- Temperature
+
+**操作按钮**：
+- 取消、测试连接、保存
+
+---
+
+## 五、数据模型
+
+### 5.1 文字稿（Transcript）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| id | string | 唯一标识 |
+| filename | string | 原始文件名 |
+| file_path | string | 服务器存储路径 |
+| line_count | int | 总行数 |
+| file_size | int | 文件大小（字节） |
+| status | enum | pending / processing / done / error |
+| uploaded_at | datetime | 上传时间 |
+
+### 5.2 讲述者（Person）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| id | string | 唯一标识 |
+| name | string | 真实姓名 |
+| nickname | string | 文字稿中的昵称（如"14班+HCM"） |
+| filename | string | 原始 DOCX 文件名 |
+| file_path | string | 服务器存储路径 |
+| photo_path | string | 提取的照片路径 |
+| info | json | 从 DOCX 提取的个人信息（年龄、性别、城市、职业等） |
+| uploaded_at | datetime | 上传时间 |
+
+### 5.3 故事（Story）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| id | string | 唯一标识 |
+| title | string | AI 生成的故事标题 |
+| summary | string | AI 生成的故事摘要 |
+| content | text | 完整故事内容（仅学员讲述部分，已过滤老师对话） |
+| speaker_nickname | string | 文字稿中的讲述者昵称（如"14班+HCM"） |
+| source_transcript_id | string | 来源文字稿 ID |
+| source_lines | json | 在原文中的行号范围 [{start, end}] |
+| duration_minutes | float | 预估时长（分钟） |
+| confidence | float | 置信度（0.0-1.0） |
+| person_id | string | 匹配的讲述者 ID（可为空） |
+| match_status | enum | pending / matched / deleted |
+| created_at | datetime | 提取时间 |
+
+### 5.4 系统配置（Config）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| llm_provider | string | API 提供商 |
+| llm_model | string | 模型名称 |
+| llm_base_url | string | API Base URL |
+| llm_api_key | string | API Key（加密存储） |
+| segment_max_lines | int | 话题分段最大行数 |
+| story_min_lines | int | 故事最小行数 |
+| confidence_threshold | float | 置信度阈值 |
+| concurrency | int | 并发请求数 |
+| temperature | float | Temperature |
+
+---
+
+## 六、输出文档模板
+
+每个讲述者生成一份 DOCX，结构如下：
+
+```
+┌─────────────────────────────────┐
+│  [照片]                          │
+│  姓名：张小花                     │
+│  性别：女  年龄：35岁             │
+│  城市：北京  职业：全职妈妈         │
+├─────────────────────────────────┤
+│                                 │
+│  她的故事                        │
+│  ─────                          │
+│  故事正文第一段...                 │
+│                                 │
+│  故事正文第二段...                 │
+│                                 │
+│  （如有多个故事，依次排列）         │
+│                                 │
+└─────────────────────────────────┘
+```
+
+- 简洁模板，不花哨
+- 个人信息在顶部，照片居中
+- 故事正文在下方（仅学员讲述内容，已过滤老师对话）
+- 一位讲述者可能有多个故事，依次排列
+
+---
+
+## 七、非功能性需求
+
+| 需求 | 说明 |
+|------|------|
+| **局域网访问** | 部署在局域网内，多人可通过浏览器访问 |
+| **扩展性** | LLM 可替换，后续可扩展文章搜索引擎 |
+| **数据安全** | API Key 加密存储，文件仅在服务器本地处理 |
+| **并发处理** | 支持多文件并发提取，避免前端长时间等待 |
+| **容错性** | 单个文件处理失败不影响其他文件 |

docs/Story-Extractor-后端架构设计文档.md

@@ -0,0 +1,757 @@
+# Story Extractor — 后端架构设计文档
+
+> 版本：v1.3 · 2026-04-21
+
+---
+
+## 一、架构总览
+
+### 1.1 技术栈
+
+| 层级 | 技术 | 选型理由 |
+|------|------|----------|
+| **Web 框架** | FastAPI | 异步支持、自动 API 文档、类型提示、高性能 |
+| **前端** | Vue 3 + Element Plus | 组件丰富、中文生态好、适合管理类交互 |
+| **数据库** | SQLite | 轻量零安装、局域网场景够用、后续可迁移 |
+| **LLM 接口** | OpenAI Python SDK（兼容协议） | 一套代码适配 OpenAI / DeepSeek / 通义千问 / Ollama |
+| **DOCX 处理** | python-docx | 读写 DOCX 的标准 Python 库 |
+| **部署** | Docker Compose | 一键启动，局域网分发方便 |
+
+> **设计原则：极简架构**，不引入 Redis、Celery、Nginx 等外部依赖，全部基于 FastAPI 原生能力。
+
+### 1.2 系统架构图
+
+```
+                    ┌──────────────────┐
+                    │   浏览器 (Vue 3)  │
+                    └────────┬─────────┘
+                             │ HTTP / SSE
+                    ┌────────┴─────────┐
+                    │                  │
+             ┌──────┴──────┐ ┌────────┴──┐
+             │  FastAPI     │ │  SQLite   │
+             │  (Web + API  │ │  (数据库)  │
+             │  + 静态托管  │ │           │
+             │  + 后台任务) │ │           │
+             └──────┬──────┘ └───────────┘
+                    │
+        ┌───────────┼───────────┐
+        │           │           │
+  ┌─────┴─────┐ ┌──┴───┐ ┌────┴────┐
+  │ 预处理器   │ │ LLM  │ │ DOCX    │
+  │ + 提取器   │ │ 服务  │ │ 生成器  │
+  └───────────┘ └──────┘ └─────────┘
+```
+
+**极简设计**：
+- ❌ 无 Redis（进度追踪用内存）
+- ❌ 无 Celery（后台任务用 asyncio）
+- ❌ 无 Nginx（FastAPI 托管静态文件）
+- ✅ **单容器部署**，一键启动
+
+### 1.3 项目目录结构
+
+```
+story-extractor/
+├── docker-compose.yml
+├── Dockerfile
+├── requirements.txt
+├── .env.example
+│
+├── backend/
+│   ├── app/
+│   │   ├── __init__.py
+│   │   ├── main.py                  # FastAPI 入口，路由注册，SSE
+│   │   ├── config.py                # 配置管理（读取 .env）
+│   │   ├── database.py              # SQLite 连接与 ORM
+│   │   ├── state.py                 # 内存状态管理（进度追踪）
+│   │   │
+│   │   ├── api/                     # API 路由层
+│   │   │   ├── __init__.py
+│   │   │   ├── files.py             # 文件上传/删除/列表
+│   │   │   ├── extraction.py        # 提取任务管理
+│   │   │   ├── stories.py           # 故事 CRUD（含删除）
+│   │   │   ├── matching.py          # 匹配关系管理
+│   │   │   ├── export.py            # 预览与导出
+│   │   │   └── settings.py          # 系统配置
+│   │   │
+│   │   ├── models/                  # 数据模型（SQLAlchemy）
+│   │   │   ├── __init__.py
+│   │   │   ├── transcript.py
+│   │   │   ├── person.py
+│   │   │   ├── story.py
+│   │   │   └── config.py
+│   │   │
+│   │   ├── schemas/                 # Pydantic 请求/响应模型
+│   │   │   ├── __init__.py
+│   │   │   ├── file.py
+│   │   │   ├── story.py
+│   │   │   ├── match.py
+│   │   │   └── config.py
+│   │   │
+│   │   └── services/                # 业务逻辑层
+│   │       ├── __init__.py
+│   │       ├── file_service.py      # 文件存储与管理
+│   │       ├── docx_parser.py       # DOCX 解析
+│   │       ├── preprocessor.py      # 预处理（过滤非故事内容）
+│   │       ├── extractor.py         # 故事提取核心（三阶段 Pipeline）
+│   │       ├── llm_client.py        # LLM 抽象层
+│   │       ├── match_service.py     # 匹配关系管理
+│   │       └── docx_generator.py    # 最终 DOCX 生成
+│   │
+│   ├── uploads/                     # 上传文件存储
+│   │   ├── transcripts/
+│   │   └── persons/
+│   │
+│   └── exports/                     # 导出文件存储
+│
+├── frontend/                        # Vue 3 前端项目
+│   ├── src/
+│   │   ├── views/
+│   │   │   ├── WelcomeView.vue
+│   │   │   ├── UploadView.vue
+│   │   │   ├── ExtractView.vue
+│   │   │   ├── MatchView.vue
+│   │   │   └── ExportView.vue
+│   │   ├── components/
+│   │   │   ├── Stepper.vue
+│   │   │   ├── BottomNav.vue
+│   │   │   ├── UploadZone.vue
+│   │   │   ├── StoryCard.vue
+│   │   │   ├── PersonCard.vue
+│   │   │   ├── MatchPanel.vue
+│   │   │   ├── DocPreview.vue
+│   │   │   ├── SettingsModal.vue
+│   │   │   └── Toast.vue
+│   │   ├── api/
+│   │   │   └── index.js
+│   │   ├── stores/
+│   │   │   └── app.js
+│   │   ├── App.vue
+│   │   └── main.js
+│   ├── package.json
+│   └── vite.config.js
+│
+├── Dockerfile
+├── docker-compose.yml
+├── requirements.txt
+└── .env.example
+```
+
+---
+
+## 二、API 设计
+
+### 2.1 API 总览
+
+所有 API 以 `/api/v1` 为前缀。
+
+| 模块 | 方法 | 路径 | 说明 |
+|------|------|------|------|
+| **文件** | POST | `/api/v1/files/transcripts` | 上传文字稿（支持批量） |
+| | POST | `/api/v1/files/persons` | 上传讲述者信息（支持批量） |
+| | GET | `/api/v1/files/transcripts` | 获取文字稿列表 |
+| | GET | `/api/v1/files/persons` | 获取讲述者列表 |
+| | DELETE | `/api/v1/files/transcripts/{id}` | 删除文字稿 |
+| | DELETE | `/api/v1/files/persons/{id}` | 删除讲述者信息 |
+| | GET | `/api/v1/files/persons/{id}/preview` | 预览讲述者信息 |
+| **提取** | POST | `/api/v1/extraction/start` | 启动提取任务（后台异步） |
+| | GET | `/api/v1/extraction/status` | 获取提取进度 |
+| | GET | `/api/v1/extraction/status/stream` | SSE 实时进度推送 |
+| | POST | `/api/v1/extraction/stop` | 停止提取任务 |
+| **故事** | GET | `/api/v1/stories` | 获取故事列表（支持筛选） |
+| | GET | `/api/v1/stories/{id}` | 获取故事详情 |
+| | PUT | `/api/v1/stories/{id}` | 编辑故事 |
+| | DELETE | `/api/v1/stories/{id}` | 删除故事 |
+| **匹配** | POST | `/api/v1/matches` | 创建匹配关系 |
+| | DELETE | `/api/v1/matches/{story_id}` | 取消匹配 |
+| | GET | `/api/v1/matches` | 获取所有匹配关系 |
+| **导出** | GET | `/api/v1/export/preview/{person_id}` | 预览文档 |
+| | GET | `/api/v1/export/download/{person_id}` | 下载单个 DOCX |
+| | GET | `/api/v1/export/download-all` | 批量下载 ZIP |
+| | GET | `/api/v1/export/list` | 获取导出列表 |
+| **设置** | GET | `/api/v1/settings` | 获取系统配置 |
+| | PUT | `/api/v1/settings` | 更新系统配置 |
+| | POST | `/api/v1/settings/test-llm` | 测试 LLM 连接 |
+
+### 2.2 关键 API 详细设计
+
+#### 启动提取任务
+
+```
+POST /api/v1/extraction/start
+```
+
+**响应**：
+```json
+{
+  "task_id": "uuid",
+  "status": "started",
+  "total_files": 10
+}
+```
+
+#### SSE 实时进度推送
+
+```
+GET /api/v1/extraction/status/stream
+```
+
+**实现方式**：FastAPI `StreamingResponse` + `asyncio.Queue`
+
+```python
+@router.get("/extraction/status/stream")
+async def stream_progress():
+    async def event_generator():
+        while True:
+            data = await extraction_state.get_update()
+            yield f"event: progress\ndata: {json.dumps(data)}\n\n"
+            if data.get("status") == "completed":
+                break
+    return StreamingResponse(event_generator(), media_type="text/event-stream")
+```
+
+**SSE 事件格式**：
+```
+event: progress
+data: {
+  "percent": 35,
+  "current_file": "2026年1月21日 第10营训练营.docx",
+  "action": "故事识别：检测学员发言...",
+  "files_read": 3,
+  "files_total": 10,
+  "segments_found": 42,
+  "stories_found": 2
+}
+
+event: story_found
+data: {
+  "id": "uuid",
+  "title": "代际传承的体罚创伤",
+  "speaker_nickname": "14班+HCM",
+  "summary": "从小被妈妈体罚，长大后用同样方式对待女儿...",
+  "confidence": 0.85,
+  "source": "2026年1月21日 第10营训练营.docx"
+}
+
+event: completed
+data: {
+  "total_files": 10,
+  "total_stories": 15,
+  "duration_seconds": 186
+}
+```
+
+#### 停止提取任务
+
+```
+POST /api/v1/extraction/stop
+```
+
+**说明**：通过 `asyncio.Event` 取消后台任务。
+
+---
+
+## 三、核心服务设计
+
+### 3.1 内存状态管理（state.py）
+
+**替代 Redis 的方案**：用 Python 模块级变量 + `asyncio.Queue` 管理进度。
+
+```python
+class ExtractionState:
+    """提取任务状态管理（内存中，无需 Redis）"""
+
+    def __init__(self):
+        self.is_running: bool = False
+        self.cancel_event: asyncio.Event = asyncio.Event()
+        self.progress: dict = {
+            "percent": 0,
+            "current_file": "",
+            "action": "",
+            "files_read": 0,
+            "files_total": 0,
+            "segments_found": 0,
+            "stories_found": 0,
+        }
+        self.queue: asyncio.Queue = asyncio.Queue()
+
+    def update(self, **kwargs):
+        """更新进度，同时推送到 SSE 队列"""
+        self.progress.update(kwargs)
+        self.queue.put_nowait(self.progress.copy())
+
+    async def get_update(self) -> dict:
+        """SSE 消费端：等待新的进度更新"""
+        return await self.queue.get()
+
+    def reset(self):
+        """重置状态"""
+        self.is_running = False
+        self.cancel_event.clear()
+        self.progress = {k: 0 if k != "" else "" for k in self.progress}
+        self.queue = asyncio.Queue()
+
+# 全局单例
+extraction_state = ExtractionState()
+```
+
+**局限性**：
+- 服务重启后进度丢失（可接受，重新提取即可）
+- 单进程内有效（FastAPI 单进程部署，无问题）
+
+### 3.2 后台任务管理
+
+**替代 Celery 的方案**：FastAPI `asyncio.create_task()`
+
+```python
+# main.py
+from app.services.extractor import run_extraction
+
+@router.post("/extraction/start")
+async def start_extraction():
+    if extraction_state.is_running:
+        return {"error": "提取任务正在进行中"}
+    extraction_state.reset()
+    extraction_state.is_running = True
+    asyncio.create_task(run_extraction(extraction_state))
+    return {"task_id": str(uuid4()), "status": "started"}
+
+@router.post("/extraction/stop")
+async def stop_extraction():
+    extraction_state.cancel_event.set()
+    return {"status": "stopping"}
+```
+
+```python
+# extractor.py
+async def run_extraction(state: ExtractionState):
+    """后台提取任务主循环"""
+    try:
+        transcripts = get_all_transcripts()
+        state.update(files_total=len(transcripts))
+
+        for i, transcript in enumerate(transcripts):
+            if state.cancel_event.is_set():
+                break
+
+            state.update(
+                current_file=transcript.filename,
+                action="预处理：过滤课堂管理内容...",
+                files_read=i + 1,
+            )
+
+            paragraphs = parse_transcript(transcript.file_path)
+            filtered = preprocess(paragraphs)
+            segments = await segment_text(filtered, state)
+            stories = await identify_and_extract(segments, state)
+
+            save_stories(stories, transcript.id)
+
+            state.update(
+                percent=round((i + 1) / len(transcripts) * 100),
+                action=f"已完成：{transcript.filename}",
+            )
+
+        state.update(status="completed", percent=100)
+    except Exception as e:
+        state.update(status="error", action=f"错误：{str(e)}")
+    finally:
+        state.is_running = False
+```
+
+### 3.3 文字稿解析器（docx_parser.py）
+
+```python
+def parse_transcript(file_path: str) -> TranscriptData:
+    """
+    解析直播回放文字稿 DOCX
+    格式：发言者(时间戳): 内容
+
+    返回：
+    - paragraphs: [{index, speaker, timestamp, content, raw_text}]
+    - total_lines: 总行数
+    """
+```
+
+**关键处理**：
+- 解析每段的发言者、时间戳、内容
+- 识别发言者类型：老师（卢慧老师、静静、督导）vs 学员
+- 建立行号映射关系
+
+### 3.4 预处理器（preprocessor.py）
+
+```python
+class Preprocessor:
+    """预处理：过滤非故事内容"""
+
+    def filter_classroom_management(self, paragraphs: list) -> list:
+        """过滤课堂管理内容（改昵称、纪律提醒、进组讨论）"""
+
+    def filter_short_responses(self, paragraphs: list) -> list:
+        """过滤学员短回应（"好的""收到"等，长度<10字）"""
+
+    def filter_healing_exercises(self, paragraphs: list) -> list:
+        """过滤疗愈练习重复内容（连续"我愿意看见且释放"模式）"""
+
+    def extract_student_segments(self, paragraphs: list) -> list:
+        """提取学员发言段落，保留上下文"""
+```
+
+**过滤规则**：
+- 课堂管理关键词：昵称、纪律、进组、会议室、扣三个一
+- 短回应：长度 < 10 字，且不含情感词
+- 疗愈练习：连续 3 段以上匹配"我愿意看见且释放"模式
+
+### 3.5 故事提取器（extractor.py）
+
+**核心算法：三阶段 Pipeline**
+
+```
+阶段1: 预处理 → 阶段2: 故事识别 → 阶段3: 故事提取
+```
+
+#### 阶段一：预处理（规则过滤）
+
+```python
+def preprocess(self, paragraphs: list) -> list:
+    """
+    1. 过滤课堂管理内容
+    2. 过滤短回应
+    3. 过滤疗愈练习重复
+    4. 提取学员发言段落
+    """
+```
+
+#### 阶段二：故事识别（LLM）
+
+**分块策略**：
+- 按 300-500 行分块，块间重叠 30 行
+- 每块送入 LLM 判断是否包含学员故事
+
+**Prompt 模板**：
+```
+你是一个文本分析助手。以下是某心理成长训练营直播的文字稿片段。
+
+每段格式为：发言者(时间戳): 内容
+
+请判断该片段是否包含"学员个人故事"。
+
+【学员个人故事】的定义：
+1. 学员（非老师）讲述的自己或家人的真实经历
+2. 涉及具体的事件、人物、时间线
+3. 有情感色彩（痛苦、挣扎、觉醒等）
+4. 主题通常围绕：原生家庭、亲子关系、婚姻情感、财富限制、身体健康、职业发展
+
+【不属于学员故事】的内容：
+1. 老师的理论讲解和方法论
+2. 课堂管理（改昵称、纪律提醒）
+3. 疗愈练习引导语（"我愿意看见且释放..."）
+4. 学员的简短回应（"好的""明白了"）
+5. 老师引用的案例（非现场学员讲述）
+
+请返回 JSON：
+{
+  "has_story": true/false,
+  "story_speaker": "学员昵称（如14班+HCM）",
+  "story_topic": "一句话概括主题（15字以内）",
+  "story_start_line": 起始行号,
+  "story_end_line": 结束行号,
+  "confidence": 0.0-1.0
+}
+```
+
+#### 阶段三：故事提取（LLM）
+
+**Prompt 模板**：
+```
+以下是某学员在训练营中讲述的个人故事（对话片段）。
+
+请完成以下任务：
+1. 整理出完整的故事叙述（去掉老师的提问和引导语，只保留学员的讲述）
+2. 为故事起一个简洁的标题（10字以内）
+3. 写一段故事摘要（80-150字，第三人称）
+4. 识别故事的核心主题标签
+
+【重要】只提取学员讲述的内容，过滤掉：
+- 老师的提问、引导、追问
+- 疗愈练习引导语
+- 其他学员的发言
+
+【输出格式】JSON：
+{
+  "title": "故事标题",
+  "summary": "故事摘要",
+  "content": "整理后的完整故事叙述（第一人称，仅学员讲述部分）",
+  "themes": ["原生家庭", "体罚", "代际传承"],
+  "emotion_tone": "痛苦/挣扎/觉醒/释然/..."
+}
+```
+
+### 3.6 LLM 客户端（llm_client.py）
+
+```python
+class LLMClient:
+    """LLM 统一调用接口"""
+
+    def __init__(self, config: LLMConfig):
+        self.client = AsyncOpenAI(
+            api_key=config.api_key,
+            base_url=config.base_url,
+        )
+        self.model = config.model
+
+    async def chat(self, messages: list[dict], temperature: float = 0.3) -> str:
+        """发送对话请求"""
+
+    async def chat_json(self, messages: list[dict], temperature: float = 0.3) -> dict:
+        """发送对话请求，期望返回 JSON"""
+```
+
+**支持的提供商**（通过 base_url 切换）：
+
+| 提供商 | Base URL |
+|--------|----------|
+| OpenAI | `https://api.openai.com/v1` |
+| DeepSeek | `https://api.deepseek.com/v1` |
+| 通义千问 | `https://dashscope.aliyuncs.com/compatible-mode/v1` |
+| Ollama (本地) | `http://localhost:11434/v1` |
+
+### 3.7 匹配服务（match_service.py）
+
+```python
+class MatchService:
+    """匹配关系管理"""
+
+    def create_match(self, story_id: str, person_id: str) -> Match:
+        """创建匹配关系"""
+
+    def remove_match(self, story_id: str) -> bool:
+        """取消匹配"""
+
+    def get_matched_persons(self) -> list[Person]:
+        """获取已匹配的讲述者列表（用于导出）"""
+```
+
+### 3.8 DOCX 生成器（docx_generator.py）
+
+```python
+def generate_person_doc(person: Person, stories: list[Story], output_path: str):
+    """
+    为一位讲述者生成最终 DOCX 文档
+
+    结构：
+    ┌─ 照片（居中）
+    ├─ 个人信息（姓名、性别、年龄、城市、职业）
+    ├─ 分隔线
+    └─ 故事正文（仅学员讲述内容，多个故事依次排列）
+    """
+```
+
+---
+
+## 四、错误处理
+
+| 错误类型 | 处理方式 |
+|----------|----------|
+| 单个文字稿解析失败 | 记录错误，跳过，继续处理其他文件 |
+| LLM API 调用失败 | 自动重试 3 次（指数退避），仍失败则标记该段为 error |
+| LLM 返回格式异常 | 尝试修复 JSON，无法修复则丢弃该段结果 |
+| DOCX 文件损坏 | 提示用户文件异常，跳过 |
+| 提取任务中断（服务重启） | 进度丢失，用户需重新启动提取 |
+
+---
+
+## 五、数据库设计
+
+### 5.1 ER 关系
+
+```
+Transcript 1───* Story *───0..1 Person
+
+Config (单例)
+```
+
+### 5.2 表结构
+
+#### transcripts 表
+
+```sql
+CREATE TABLE transcripts (
+    id TEXT PRIMARY KEY,
+    filename TEXT NOT NULL,
+    file_path TEXT NOT NULL,
+    line_count INTEGER DEFAULT 0,
+    file_size INTEGER DEFAULT 0,
+    status TEXT DEFAULT 'pending',
+    error_message TEXT,
+    uploaded_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+```
+
+#### persons 表
+
+```sql
+CREATE TABLE persons (
+    id TEXT PRIMARY KEY,
+    name TEXT NOT NULL,
+    nickname TEXT,
+    filename TEXT NOT NULL,
+    file_path TEXT NOT NULL,
+    photo_path TEXT,
+    info TEXT,
+    uploaded_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
+);
+```
+
+#### stories 表
+
+```sql
+CREATE TABLE stories (
+    id TEXT PRIMARY KEY,
+    title TEXT NOT NULL,
+    summary TEXT,
+    content TEXT,
+    speaker_nickname TEXT,
+    source_transcript_id TEXT NOT NULL,
+    source_lines TEXT,
+    duration_minutes REAL,
+    confidence REAL,
+    confidence_level TEXT,
+    person_id TEXT,
+    match_status TEXT DEFAULT 'pending',
+    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+    FOREIGN KEY (source_transcript_id) REFERENCES transcripts(id),
+    FOREIGN KEY (person_id) REFERENCES persons(id)
+);
+```
+
+#### config 表
+
+```sql
+CREATE TABLE config (
+    key TEXT PRIMARY KEY,
+    value TEXT NOT NULL
+);
+```
+
+---
+
+## 六、部署方案
+
+### 6.1 Docker Compose（单容器）
+
+```yaml
+version: '3.8'
+services:
+  app:
+    build: .
+    ports:
+      - "80:8000"
+    volumes:
+      - ./data/uploads:/app/uploads
+      - ./data/exports:/app/exports
+      - ./data/db:/app/data
+    env_file: .env
+```
+
+### 6.2 FastAPI 托管静态文件
+
+```python
+# main.py
+from fastapi import FastAPI
+from fastapi.staticfiles import StaticFiles
+
+app = FastAPI()
+
+# API 路由（先注册）
+app.include_router(files_router, prefix="/api/v1")
+app.include_router(extraction_router, prefix="/api/v1")
+app.include_router(stories_router, prefix="/api/v1")
+app.include_router(matching_router, prefix="/api/v1")
+app.include_router(export_router, prefix="/api/v1")
+app.include_router(settings_router, prefix="/api/v1")
+
+# 托管前端静态文件（放在最后，作为 fallback）
+app.mount("/", StaticFiles(directory="frontend/dist", html=True), name="static")
+```
+
+### 6.3 环境变量（.env）
+
+```env
+# LLM 配置
+LLM_PROVIDER=openai
+LLM_MODEL=gpt-4
+LLM_BASE_URL=https://api.openai.com/v1
+LLM_API_KEY=sk-xxx
+
+# 提取参数
+SEGMENT_MAX_LINES=500
+STORY_MIN_LINES=50
+CONFIDENCE_THRESHOLD=0.5
+TEMPERATURE=0.3
+
+# 应用
+APP_HOST=0.0.0.0
+APP_PORT=8000
+UPLOAD_DIR=/app/uploads
+EXPORT_DIR=/app/exports
+```
+
+### 6.4 启动流程
+
+```bash
+# 1. 配置环境变量
+cp .env.example .env
+vim .env
+
+# 2. 构建前端
+cd frontend && npm install && npm run build && cd ..
+
+# 3. 一键启动（单容器）
+docker-compose up -d
+
+# 4. 访问
+# http://<局域网IP>
+```
+
+### 6.5 依赖清单（requirements.txt）
+
+```
+fastapi>=0.110.0
+uvicorn>=0.27.0
+python-multipart>=0.0.9
+sqlalchemy>=2.0
+aiosqlite>=0.19.0
+python-docx>=1.1.0
+openai>=1.12.0
+pydantic>=2.0
+```
+
+> 仅 8 个依赖，无 Redis、无 Celery。
+
+---
+
+## 七、扩展性设计
+
+### 7.1 LLM 升级
+
+- 所有 LLM 调用通过 `LLMClient` 抽象层
+- 切换模型只需修改配置（base_url + model + api_key）
+
+### 7.2 文章搜索引擎（预留）
+
+- stories 表已存储完整内容 + 结构化元数据
+- 后续可接入 SQLite FTS5 全文搜索或 Meilisearch
+
+### 7.3 多用户支持（预留）
+
+- 当前为单用户模式（局域网共享）
+- 后续可加入用户认证（JWT）
+
+### 7.4 从极简架构升级
+
+如果后续需要更强的任务管理能力，可以平滑升级：
+- `ExtractionState` → Redis
+- `asyncio.create_task()` → Celery
+- 架构设计已预留接口，业务逻辑层不需要改动

docs/extraction-design.md

@@ -0,0 +1,295 @@
+# 故事提取工具 - 提取逻辑设计文档 v4
+
+## 一、目标
+
+**输入**：录像转文字稿（回溯疗愈课堂的对话记录，一个或多个 DOCX 文件）
+
+**输出**：每个学员一个结构化故事，包含 5 个部分：
+1. 讲述者简介
+2. 当下的困境
+3. 过往的故事
+4. 转变
+5. 原文金句
+
+**核心原则**：把案主回溯的内容和面临的议题提取整理成第三人称故事。
+
+---
+
+## 二、整体流程
+
+```
+多个文档 → 按顺序逐个处理 → 每个文档独立输出故事 → 汇总展示
+```
+
+- 多个文档按顺序处理，一个文档处理完再处理下一个（避免 API 限流）
+- 每个文档内部：代码预处理 → 并行判断/提取 → 并行切片提取 → 重组 → 并行生成故事
+- **核心优化**：所有 LLM 调用均使用 `asyncio.gather` 并行执行，总耗时 ≈ 最慢的单次调用耗时，而非所有调用耗时之和
+
+---
+
+## 三、单文档处理流程（核心）
+
+### 第一步：代码预处理（不用 LLM）
+
+**目的**：把原始对话记录拆分成"以学员为单位"的段落，每个段落包含该学员从第一次发言到最后一次发言之间的**所有内容**。
+
+**不做的事情**：
+- 不过滤课堂管理内容（实测过滤掉的行很少，且关键词硬编码不通用，不值得做）
+
+**具体逻辑**：
+
+#### 1.1 统计说话人
+
+遍历所有行，统计每个说话人的：
+- 发言行数
+- 总字数
+- 第一次出现的行号（first）
+- 最后一次出现的行号（last）
+
+#### 1.2 识别老师
+
+- 发言行数最多的人就是老师
+- 不依赖硬编码名字，适用于任何课程
+
+#### 1.3 过滤杂音
+
+- 老师本身不参与拆分
+- 发言 < 20 句的说话人视为杂音，直接跳过（通常是主持人、助教、偶尔插话的人）
+
+#### 1.4 按时间范围截取（核心逻辑）
+
+对每个有效学员，从该学员**第一次发言的行**到**最后一次发言的行**，中间**所有行全部截取**，形成一个完整段落。
+
+**关键特性：不同学员的段落之间允许重叠。**
+
+举例说明：
+```
+原文行号：  131  132  ...  167  168  ...  321  322  ...  429  430  ...  599  600  ...  800
+说话人：    04   02   ...  05   02   ...  06   02   ...  04   02   ...  05   02   ...  07
+
+说话人04段落：截取行[131-429]，包含04、02(老师)、05、06的所有对话
+说话人05段落：截取行[167-599]，包含05、02(老师)、06、04的部分对话
+说话人07段落：截取行[603-800]，包含07、02(老师)的对话
+```
+
+**为什么允许重叠？**
+- 课堂场景中，多个学员可能围绕同一个主题互动（比如说话人04和说话人05都在讨论高考漏题创伤）
+- 老师在不同学员之间的引导是连贯的，不应该被割裂
+- 重叠部分的老师对话对两个学员的故事都有价值，应该分别保留
+
+#### 1.5 过滤
+
+对截取后的段落进行两轮过滤：
+
+**过滤条件 A：短段落**
+- 学员发言 < 15 句 → 丢弃（不太可能有故事）
+- 学员总字数 < 80 字 → 丢弃
+
+**过滤条件 B：非主讲述人**
+- 当事人发言行数 / 段落总行数 < 10% → 丢弃
+- 含义：截取了很大一段范围，但里面绝大部分话不是该学员说的，说明该学员只是旁观者或偶尔插话，不是主讲述人
+- 举例：说话人03截取了 658 行，但其中只有 21 行是03说的（占比 3.2%），说明03在整个课堂期间只是偶尔插话，不是主讲述人
+
+#### 1.6 输出
+
+- 每个通过过滤的学员一个段落
+- 段落包含该学员时间范围内的所有对话（老师、其他学员、本人）
+- 段落按学员第一次出现的时间排序
+
+---
+
+### 第二步：判断故事 + 提取事实（并行执行）
+
+**目的**：根据段落长度走不同路径，短段落省一次 LLM 调用。**所有段落并行处理，互不阻塞。**
+
+**并行策略**：使用 `asyncio.gather` 同时处理所有段落，总耗时 ≈ 最慢那个段落的处理时间。
+
+**短段落（≤ 4000 字）**：
+- 跳过"判断故事"步骤，直接让 LLM 提取事实
+- 理由：短段落内容少，提取失败的成本低，省一次判断调用更划算
+
+**长段落（> 4000 字）**：
+- 先让 LLM 判断"是否包含个人故事"
+- 输出：`{is_story: true/false, confidence: 0.0~1.0, story_hints: "一句话描述"}`
+- confidence < 0.5 或 is_story = false → 跳过，不浪费后续切片提取的调用
+- 有故事 → 进入第三步切片提取
+
+---
+
+### 第三步：切片提取事实（Map 阶段，仅长段落执行）
+
+**目的**：对确认有故事的长段落，切成小块并行提取关键信息，避免输出截断。
+
+**为什么拆成"提取事实"和"生成故事"两步？**
+- 直接让 LLM 从长文本写故事 → 输出太长，必然截断
+- 先提取事实（输出短，不会截断）→ 再基于短素材写故事（输入短，输出可控）
+
+**切片策略**：
+- 以自然段为最小单位，累加到约 3000 字时，在最近的段落结尾处切断
+- 块之间重叠 **5 个自然段**（避免信息断裂）
+- 短段落（≤ 4000 字）在第二步已直接提取，不会进入此步
+
+**每块的提取任务**（并行执行）：
+- 输入：一块对话文本 + 该学员的故事主题（来自第二步的 hints）
+- 输出：结构化事实
+  ```json
+  {
+    "events": ["事件1（尽量详细）", "事件2"],
+    "emotions": ["情感1", "情感2"],
+    "quotes": ["讲述者原话1", "讲述者原话2", "讲述者原话3"],
+    "key_people": ["相关人物1", "相关人物2"]
+  }
+  ```
+- **关键约束**：每块只提取事实，不做写作。尽量详细、尽量多提取，保留具体细节
+
+---
+
+### 第四步：重组（Reduce 阶段，纯代码）
+
+**目的**：把多个切片的提取结果合并成完整的素材。
+
+- 输入：同一学员所有切片的提取结果
+- 处理：
+  - 事件：按原文出现顺序排列，去除重复
+  - 情感：去重
+  - 原话：去重（完全相同的去掉），不限制数量，不限制长度
+  - 人物：去重
+- 输出：完整素材
+  ```json
+  {
+    "events": ["按时间排序的事件列表"],
+    "emotions": ["去重后的情感列表"],
+    "quotes": ["去重后的原话列表（全部保留）"],
+    "key_people": ["去重后的相关人物列表"],
+    "story_hints": "故事主题"
+  }
+  ```
+
+---
+
+### 第五步：生成最终故事（并行执行）
+
+**目的**：基于重组后的素材，生成结构化的第三人称故事。**所有故事并行生成，互不阻塞。**
+
+**并行策略**：使用 `asyncio.gather` 同时生成所有故事，总耗时 ≈ 最慢那个故事的生成时间。
+
+- 输入：重组后的完整素材（短文本，几百字）
+- 输出：最终故事 JSON
+  ```json
+  {
+    "title": "标题",
+    "summary": "80字摘要",
+    "tags": ["标签1", "标签2"],
+    "speaker_nickname": "讲述者昵称",
+    "content": {
+      "讲述者": "讲述者介绍（不限制长度，充分展开）",
+      "当下的困境": "当前面临的困境（不限制长度，充分展开）",
+      "过往的故事": "过去的经历故事（不限制长度，充分展开）",
+      "转变": "疗愈过程中的转变（不限制长度，充分展开）",
+      "原文金句": ["原话1", "原话2", "原话3"]
+    }
+  }
+  ```
+- **关键**：LLM 的输入是"已提取好的素材"（几百字），不是原始长文本
+- LLM 的任务是"基于素材写作"，不需要再从原文中提取信息
+- **不限制输出长度**，要求素材中的具体经历、事件细节、情感变化尽量保留，让故事丰满
+- **原文金句为数组格式**，避免字符串中的引号导致 JSON 解析失败
+
+---
+
+## 四、数据流图
+
+```
+原始 DOCX 文件
+    │
+    ▼
+[代码预处理]
+    │  1. 统计说话人（行数、字数、首末行号）
+    │  2. 识别老师（发言最多的人）
+    │  3. 过滤杂音（< 20句）
+    │  4. 按时间范围截取（第一句到最后一句，允许重叠）
+    │  5. 过滤短段落（<15句 / <80字）
+    │  6. 过滤非主讲述人（当事人占比 < 10%）
+    ▼
+学员段落列表: [学员A段落, 学员B段落, 学员C段落, ...]
+    │
+    ▼
+[第二步：并行判断 + 提取] ── asyncio.gather 并行处理所有段落
+    │  ├─ 短段落(≤4000字)：直接提取事实（省1次调用）
+    │  └─ 长段落(>4000字)：先判断是否有故事 → 有故事则进入第三步
+    │  ⏱ 总耗时 ≈ 最慢那个段落的处理时间
+    │
+    ▼ (仅长段落且有故事)
+[第三步：并行切片提取] ── asyncio.gather 并行处理所有切片
+    │  按3000字切块，以自然段为单位，重叠5个自然段
+    │  ⏱ 总耗时 ≈ 最慢那个切片的提取时间
+    │
+    ▼
+[第四步：代码重组] ── 去重、排序、合并（纯代码，无需 LLM）
+    │
+    ▼
+[第五步：并行生成故事] ── asyncio.gather 并行生成所有故事
+    │  基于素材生成5部分结构化故事
+    │  ⏱ 总耗时 ≈ 最慢那个故事的生成时间
+    │
+    ▼
+最终故事列表
+```
+
+---
+
+## 五、泛化性设计
+
+| 环节 | 设计 | 为什么通用 |
+|------|------|-----------|
+| 识别老师 | 统计发言行数，最多的就是老师 | 不依赖硬编码名字 |
+| 过滤杂音 | 发言 < 20句直接跳过 | 适应各种偶尔插话的人 |
+| 按时间范围截取 | 从第一句到最后一句，中间所有内容 | 保留完整上下文，不丢失老师对话 |
+| 允许重叠 | 不同学员段落可共享内容 | 课堂中多学员讨论同一主题时，各自保留完整对话 |
+| 过滤非主讲述人 | 当事人占比 < 10% 丢弃 | 自动识别旁观者/插话者，只保留主讲述人 |
+| 切片 | 3000字一块，5个自然段重叠 | 适应任意长度 |
+
+---
+
+## 六、LLM 调用清单
+
+| 步骤 | 调用次数 | 并行方式 | 输入大小 | 输出大小 | 用途 |
+|------|---------|---------|---------|---------|------|
+| 第二步：短段落提取 | 短段落各 1 次 | **并行**（asyncio.gather） | ≤4000 字 | JSON | 直接提取事实 |
+| 第二步：长段落判断 | 长段落各 1 次 | **并行**（asyncio.gather） | ≤4000 字 | ~100 字 JSON | 过滤非故事段落 |
+| 第三步：切片提取 | 每块 1 次 | **并行**（asyncio.gather） | ≤3000 字 | JSON | 提取事件/情感/原话 |
+| 第五步：生成故事 | 每个故事 1 次 | **并行**（asyncio.gather） | 素材 ≤1000 字 | JSON | 生成最终故事 |
+
+**典型场景**：一个文档有 4 个学员段落（2短2长），其中 3 个有故事：
+- 第二步：2次短段落提取 + 2次长段落判断 = 4 次（**并行，耗时 ≈ 1次调用时间**）
+- 第三步：假设2个长段落都有故事，各切2块 = 4 次（**并行，耗时 ≈ 1次调用时间**）
+- 第五步：3 次（**并行，耗时 ≈ 1次调用时间**）
+- **总计：11 次 LLM 调用**
+
+### 耗时对比
+
+| 方案 | 第二步 | 第三步 | 第五步 | 总耗时估算 |
+|------|--------|--------|--------|-----------|
+| 串行（v3） | ~4次 × 30s = 120s | ~4次 × 30s = 120s | ~3次 × 60s = 180s | **~7 分钟** |
+| 并行（v4） | ~30s（并行） | ~30s（并行） | ~60s（并行） | **~2 分钟** |
+
+> 注：以上为单次调用约 30-60s 的粗略估算，实际耗时取决于模型响应速度和网络延迟。并行化后总耗时从分钟级降至约 2-3 分钟。
+
+---
+
+## 七、变更记录
+
+| 变更 | v2 | v3 | v4 | 原因 |
+|------|----|----|----|------|
+| 拆分方式 | 按学员拆分 + 同名合并 | 按时间范围截取（第一句到最后一句） | 不变 | v2 会丢失段落间的老师对话；v3 保留完整上下文 |
+| 段落互斥性 | 互斥（每个对话行只属于一个段落） | 允许重叠（同一行可出现在多个段落） | 不变 | 课堂中多学员讨论同一主题，老师对话对双方都有价值 |
+| 杂音过滤 | 无（依赖短段落过滤） | 发言 < 20句直接跳过 | 不变 | 提前排除主持人、助教等偶尔插话的人 |
+| 非主讲述人过滤 | 无 | 当事人占比 < 10% 丢弃 | 不变 | 自动识别旁观者，只保留真正有故事的学员 |
+| 判断故事 | 所有段落都先判断 | 短段落(≤4000字)跳过判断直接提取 | 不变 | 省一次 LLM 调用 |
+| 切片重叠 | 200字重叠 | 5个自然段重叠 | 不变 | 按自然段重叠更合理，不会在句子中间断开 |
+| 过滤管理行 | 不做 | 不做 | 不变 | 过滤掉的行很少，关键词不通用 |
+| 识别老师 | 统计发言行数 | 统计发言行数（不变） | 不变 | 通用性 |
+| 提取故事 | 先提取事实，再基于素材写故事 | 先提取事实，再基于素材写故事（不变） | 不变 | 避免输出截断 |
+| **LLM 调用方式** | 串行 | 串行 | **全并行（asyncio.gather）** | 总耗时从 ~7 分钟降至 ~2 分钟 |
+| **输出长度限制** | 有字数限制 | 有字数限制 | **不限制** | 故事更丰满，保留更多细节 |
+| **原文金句格式** | 字符串 | 字符串 | **数组** | 避免引号导致 JSON 解析失败 |

docs/故事提取工具-方案设计.md

@@ -0,0 +1,167 @@
+# 直播回放文字稿故事提取工具 — 方案设计文档
+
+## 一、项目概述
+
+### 1.1 背景
+
+手头有大量直播回放导出的文字稿（DOCX 格式），其中散落着一些个人生活故事（家长里短类），需要用 AI 自动提取整理，并与讲述者个人信息匹配，最终输出为独立的 DOCX 文档。
+
+### 1.2 目标
+
+构建一个 **Web 端可复用工具**，支持：
+- 批量上传和管理文字稿文件
+- AI 自动提取故事（过滤讲课/知识点内容）
+- 人工确认故事与讲述者的匹配关系
+- 为每个讲述者生成独立 DOCX（个人信息 + 照片 + 故事正文）
+
+### 1.3 使用场景
+
+- **部署环境：** 局域网
+- **用户规模：** 多人使用
+- **数据规模：** 20+ 份文字稿，每份上万行
+
+---
+
+## 二、需求分析
+
+### 2.1 输入
+
+| 输入类型 | 格式 | 说明 |
+|----------|------|------|
+| 直播回放文字稿 | DOCX | 每份上万行，口语化，混合讲课/知识点内容，无明确说话人标记 |
+| 讲述者个人信息 | DOCX（一人一个） | 包含姓名、照片等个人信息 |
+
+### 2.2 处理要求
+
+1. **故事提取：** 从文字稿中识别并提取个人生活故事（如家庭、情感、生活等），过滤掉纯讲课/知识点内容
+2. **长文本切分：** 上万行文字稿需智能切分，确保不截断完整故事（每个故事可能跨半小时到一小时，稀疏分布）
+3. **匹配确认：** AI 提取故事后，由人工在界面上确认每个故事对应的讲述者
+4. **文档生成：** 按简洁模板生成最终 DOCX
+
+### 2.3 输出
+
+每个讲述者一份 DOCX 文档，结构为：
+```
+┌─────────────────────────┐
+│     个人信息（含照片）     │
+├─────────────────────────┤
+│                         │
+│       故事正文           │
+│                         │
+└─────────────────────────┘
+```
+
+### 2.4 扩展性要求
+
+- LLM 可替换（支持升级/切换不同模型）
+- 后续可扩展文章搜索引擎等功能
+
+---
+
+## 三、系统架构
+
+```
+┌─────────────────────────────────────────────────┐
+│                  浏览器前端                        │
+│  ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
+│  │ 文件管理  │ │ 故事提取  │ │ 匹配确认 & 导出  │ │
+│  └──────────┘ └──────────┘ └──────────────────┘ │
+└────────────────────┬────────────────────────────┘
+                     │ HTTP API
+┌────────────────────┴────────────────────────────┐
+│              Python 后端 (FastAPI)                │
+│  ┌──────────┐ ┌──────────┐ ┌──────────────────┐ │
+│  │ 文件处理  │ │ LLM 服务  │ │ DOCX 生成       │ │
+│  └──────────┘ └──────────┘ └──────────────────┘ │
+│  ┌──────────┐ ┌──────────┐                      │
+│  │ 任务队列  │ │ 数据存储  │                      │
+│  └──────────┘ └──────────┘                      │
+└─────────────────────────────────────────────────┘
+```
+
+---
+
+## 四、用户交互流程
+
+### Step 1：上传文件
+- 上传文字稿 DOCX（支持批量，显示上传列表）
+- 上传讲述者信息 DOCX（支持批量，显示上传列表 + 预览）
+
+### Step 2：提取故事
+- 选择要处理的文字稿（可多选）
+- 点击"开始提取"，显示进度
+- LLM 实时返回提取结果
+- 展示故事列表（摘要 + 原文定位）
+
+### Step 3：匹配讲述者
+- 左侧：故事列表
+- 右侧：讲述者列表
+- 拖拽/下拉匹配（或 AI 推荐候选）
+- 确认匹配关系
+
+### Step 4：预览 & 导出
+- 预览最终 DOCX 效果
+- 单个导出 / 批量导出
+- 下载 ZIP 包
+
+---
+
+## 五、技术栈
+
+| 层级 | 技术 | 理由 |
+|------|------|------|
+| **前端** | Vue 3 + Element Plus | 组件丰富、中文生态好、拖拽匹配等交互容易实现 |
+| **后端** | FastAPI | 异步支持好、自带 API 文档、适合 AI 任务 |
+| **LLM** | 抽象接口层，先支持 OpenAI 兼容 API | 后续可无缝切换国产模型/本地模型 |
+| **DOCX 处理** | python-docx | 读写 DOCX 的标准库 |
+| **任务队列** | Celery + Redis（或内置 asyncio） | 20+ 文件处理需要异步，避免前端卡死 |
+| **数据存储** | SQLite | 轻量、无需额外安装、局域网够用 |
+| **部署** | Docker Compose | 一键启动，局域网分发方便 |
+
+---
+
+## 六、核心技术难点与解决方案
+
+| 难点 | 解决方案 |
+|------|----------|
+| **长文本切分不截断故事** | 两阶段策略：先用 LLM 做"话题分段"（识别话题边界），再对每个话题段判断是否包含故事 |
+| **故事 vs 知识点区分** | Prompt 工程 + Few-shot 示例，明确"个人生活故事"的定义和排除规则 |
+| **无说话人标记** | 提取故事后展示摘要，由人工在 GUI 上匹配讲述者；后续可加入 AI 推荐候选 |
+| **上万行文本的 LLM 上下文** | 话题分段后，每个段落在上下文窗口内处理；超长段落再按语义子段切分 |
+
+---
+
+## 七、项目结构
+
+```
+story-extractor/
+├── frontend/                    # Vue 3 前端
+│   ├── src/
+│   │   ├── views/               # 页面：文件管理、故事提取、匹配确认、导出
+│   │   ├── components/          # 通用组件
+│   │   └── api/                 # 后端 API 调用
+│   └── ...
+├── backend/                     # FastAPI 后端
+│   ├── app/
+│   │   ├── api/                 # API 路由
+│   │   ├── services/            # 业务逻辑
+│   │   │   ├── extractor.py     # 故事提取核心
+│   │   │   ├── llm.py           # LLM 抽象层
+│   │   │   ├── docx_parser.py   # DOCX 解析
+│   │   │   └── docx_generator.py # DOCX 生成
+│   │   ├── models/              # 数据模型
+│   │   └── main.py
+│   └── ...
+├── docker-compose.yml
+└── README.md
+```
+
+---
+
+## 八、待确认事项
+
+- [ ] LLM API 选型（OpenAI / 国产模型 / 本地模型）
+- [ ] 技术栈最终确认
+- [ ] 开发节奏（先跑通核心流程 vs 逐步完善）
+- [ ] 讲述者信息 DOCX 的具体字段结构（需看样例）
+- [ ] 文字稿样例（需看实际内容以优化切分和提取策略）

docs/说话人04_对比分析报告.txt

@@ -0,0 +1,133 @@
+说话人04截取文件 vs 原文对比分析报告
+==========================================
+
+一、数据汇总
+-----------
+
+1. 原始 DOCX 文件总行数: 824行
+2. 说话人04在原文中总行数: 84行 (原文行131-429)
+3. 截取文件总行数: 167行
+   - 说话人04: 84行
+   - 说话人02(老师): 83行
+
+4. 原文说话人04范围内(行131-429)的所有说话人:
+   - 说话人02(老师): 147行
+   - 说话人04: 84行
+   - 说话人05: 54行
+   - 说话人03: 8行
+   - 说话人06: 6行
+   - 总计: 299行
+
+5. 截取文件中缺失的行: 121行
+   - 老师行: 61行
+   - 其他说话人行: 60行 (说话人05: 54行, 说话人03: 8行, 说话人06: 6行)
+   - 说话人04行: 0行 (全部保留)
+
+
+二、说话人04的内容是否完整？
+-----------------------------
+
+说话人04的84行内容在截取文件中 **全部保留**，没有任何丢失。
+merge 逻辑对学员本人的内容没有造成丢失。
+
+
+三、老师在说话人04发言期间的内容是否完整？
+-------------------------------------------
+
+老师在说话人04范围内共有147行，截取文件中只保留了83行。
+**丢失了61行老师的对话。**
+
+丢失的老师对话主要分布在说话人04的12个段落之间的"中间段落"中。
+这些中间段落是其他学员(说话人03、说话人05、说话人06)发言时，
+老师与他们的互动对话。
+
+
+四、merge 逻辑丢失内容的根本原因
+---------------------------------
+
+_merge_by_speaker 的工作流程:
+  1. _split_by_student 按学员拆分: 遇到不同学员就断开
+  2. 每个段落只包含: 当前学员的行 + 老师的行
+  3. _merge_by_speaker 把同一个学员的多段 extend 合并
+
+关键问题:
+  当说话人04的段落A和段落B之间夹了其他学员(如说话人05)的段落时:
+  - 说话人05的段落中包含老师与说话人05的互动对话
+  - 这些老师对话被归入说话人05的段落(而不是说话人04的段落)
+  - merge时只合并说话人04的段落，说话人05段落中的老师对话就丢失了
+
+  但更关键的是: 说话人05在说话人04的发言期间插话，
+  老师回应说话人05的这些对话，实际上可能是在延续与说话人04的话题
+  (比如说话人05也在讨论高考漏题事件，与说话人04的故事直接相关)。
+  这些内容在merge后完全丢失了。
+
+
+五、具体丢失的重要内容
+----------------------
+
+以下是说话人04段落之间夹的中间段落中丢失的老师对话摘要:
+
+1. 段落2和段落4之间(说话人03段落):
+   - 老师安排下一位连线(行140)
+
+2. 段落4和段落7之间(说话人03+说话人05段落):
+   - 老师解释"且释放"的深层含义(行166)
+   - 老师提醒"且释放"有暗语(行168)
+
+3. 段落7和段落9之间(说话人03段落):
+   - 老师引导关于孩子高考的觉察(行173-174)
+   *** 重要: 这是老师对说话人04议题的深入引导，直接丢失！***
+
+4. 段落9和段落11之间(说话人03段落):
+   - 老师讲解教育系统的不公平和"漏中漏"概念(行224, 226)
+   *** 重要: 这是对说话人04故事的深化分析！***
+
+5. 段落11和段落14之间(说话人03+说话人05段落):
+   - 老师引导向内探索(行233, 235)
+   - 老师安排有觉知的呼吸(行237)
+
+6. 段落14和段落16之间(说话人05段落) -- **最大丢失**:
+   - 19行老师对话，包括:
+     - 老师引导说话人05表达"这不公平"的情绪(行282-296)
+     - 老师引导说话人05做愤怒释放(行298-318)
+     - 老师揭示"陪跑"真相(行318)
+   *** 非常重要: 说话人05也在处理高考漏题创伤，与说话人04同主题！***
+
+7. 段落16和段落18之间(说话人06段落):
+   - 老师解释旁观者角度(行322)
+
+8. 段落18和段落20之间(说话人05段落):
+   - 老师讲解下海捞明珠的比喻(行335)
+
+9. 段落20和段落22之间(说话人05段落):
+   - 老师引导表达情绪(行349)
+
+10. 段落24和段落30之间(说话人05+说话人03+说话人06段落) -- **第二大丢失**:
+    - 33行老师对话，包括:
+      - 老师引导对父亲的指责释放(行359-387)
+      - 老师引导"我没考过"信念释放(行400-426)
+      - 老师揭示"考的是权力不是成绩"的真相(行408, 422, 424, 426)
+    *** 极其重要: 这是整个高考创伤疗愈的高潮部分！***
+
+
+六、结论
+-------
+
+1. 说话人04本人的84行内容 **没有丢失**，全部保留在截取文件中。
+
+2. 老师在说话人04发言期间的147行中，只保留了83行(56%)，
+   丢失了61行(44%)。
+
+3. 丢失的61行老师对话中，大部分是在说话人04的段落之间，
+   老师与其他学员(主要是说话人05)互动时产生的。
+
+4. **最严重的问题**: 说话人05与说话人04讨论的是同一个主题
+   (高考漏题创伤)，老师在说话人05段落中的引导
+   (特别是"我没考过"信念释放和"考的是权力不是成绩"的真相揭示)
+   实际上是说话人04故事的延续和深化，但这些内容全部丢失了。
+
+5. merge 逻辑的设计缺陷:
+   - _split_by_student 严格按学员拆分，导致同一主题的对话被割裂
+   - _merge_by_speaker 只合并同一学员的段落，无法恢复跨学员的主题连贯性
+   - 建议: 在 merge 时，应该考虑主题连贯性，将与当前学员主题相关的
+     其他学员段落中的老师对话也纳入合并范围。

frontend/.gitignore

@@ -0,0 +1,24 @@
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+lerna-debug.log*
+
+node_modules
+dist
+dist-ssr
+*.local
+
+# Editor directories and files
+.vscode/*
+!.vscode/extensions.json
+.idea
+.DS_Store
+*.suo
+*.ntvs*
+*.njsproj
+*.sln
+*.sw?

frontend/README.md

@@ -0,0 +1,5 @@
+# Vue 3 + Vite
+
+This template should help get you started developing with Vue 3 in Vite. The template uses Vue 3 `<script setup>` SFCs, check out the [script setup docs](https://v3.vuejs.org/api/sfc-script-setup.html#sfc-script-setup) to learn more.
+
+Learn more about IDE Support for Vue in the [Vue Docs Scaling up Guide](https://vuejs.org/guide/scaling-up/tooling.html#ide-support).

frontend/index.html

@@ -0,0 +1,15 @@
+<!doctype html>
+<html lang="zh-CN">
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>Story Extractor - 直播文字稿故事提取工具</title>
+    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&display=swap" rel="stylesheet">
+    <link href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css" rel="stylesheet">
+  </head>
+  <body>
+    <div id="app"></div>
+    <script type="module" src="/src/main.js"></script>
+  </body>
+</html>

frontend/package.json

@@ -0,0 +1,20 @@
+{
+  "name": "frontend",
+  "private": true,
+  "version": "0.0.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "vite build",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "axios": "^1.15.1",
+    "vue": "^3.5.32",
+    "vue-router": "^4.6.4"
+  },
+  "devDependencies": {
+    "@vitejs/plugin-vue": "^6.0.6",
+    "vite": "^8.0.9"
+  }
+}

frontend/public/icons.svg

@@ -0,0 +1,24 @@
+<svg xmlns="http://www.w3.org/2000/svg">
+  <symbol id="bluesky-icon" viewBox="0 0 16 17">
+    <g clip-path="url(#bluesky-clip)"><path fill="#08060d" d="M7.75 7.735c-.693-1.348-2.58-3.86-4.334-5.097-1.68-1.187-2.32-.981-2.74-.79C.188 2.065.1 2.812.1 3.251s.241 3.602.398 4.13c.52 1.744 2.367 2.333 4.07 2.145-2.495.37-4.71 1.278-1.805 4.512 3.196 3.309 4.38-.71 4.987-2.746.608 2.036 1.307 5.91 4.93 2.746 2.72-2.746.747-4.143-1.747-4.512 1.702.189 3.55-.4 4.07-2.145.156-.528.397-3.691.397-4.13s-.088-1.186-.575-1.406c-.42-.19-1.06-.395-2.741.79-1.755 1.24-3.64 3.752-4.334 5.099"/></g>
+    <defs><clipPath id="bluesky-clip"><path fill="#fff" d="M.1.85h15.3v15.3H.1z"/></clipPath></defs>
+  </symbol>
+  <symbol id="discord-icon" viewBox="0 0 20 19">
+    <path fill="#08060d" d="M16.224 3.768a14.5 14.5 0 0 0-3.67-1.153c-.158.286-.343.67-.47.976a13.5 13.5 0 0 0-4.067 0c-.128-.306-.317-.69-.476-.976A14.4 14.4 0 0 0 3.868 3.77C1.546 7.28.916 10.703 1.231 14.077a14.7 14.7 0 0 0 4.5 2.306q.545-.748.965-1.587a9.5 9.5 0 0 1-1.518-.74q.191-.14.372-.293c2.927 1.369 6.107 1.369 8.999 0q.183.152.372.294-.723.437-1.52.74.418.838.963 1.588a14.6 14.6 0 0 0 4.504-2.308c.37-3.911-.63-7.302-2.644-10.309m-9.13 8.234c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.894 0 1.614.82 1.599 1.82.001 1-.705 1.82-1.6 1.82m5.91 0c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.893 0 1.614.82 1.599 1.82 0 1-.706 1.82-1.6 1.82"/>
+  </symbol>
+  <symbol id="documentation-icon" viewBox="0 0 21 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="m15.5 13.333 1.533 1.322c.645.555.967.833.967 1.178s-.322.623-.967 1.179L15.5 18.333m-3.333-5-1.534 1.322c-.644.555-.966.833-.966 1.178s.322.623.966 1.179l1.534 1.321"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M17.167 10.836v-4.32c0-1.41 0-2.117-.224-2.68-.359-.906-1.118-1.621-2.08-1.96-.599-.21-1.349-.21-2.848-.21-2.623 0-3.935 0-4.983.369-1.684.591-3.013 1.842-3.641 3.428C3 6.449 3 7.684 3 10.154v2.122c0 2.558 0 3.838.706 4.726q.306.383.713.671c.76.536 1.79.64 3.581.66"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M3 10a2.78 2.78 0 0 1 2.778-2.778c.555 0 1.209.097 1.748-.047.48-.129.854-.503.982-.982.145-.54.048-1.194.048-1.749a2.78 2.78 0 0 1 2.777-2.777"/>
+  </symbol>
+  <symbol id="github-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M9.356 1.85C5.05 1.85 1.57 5.356 1.57 9.694a7.84 7.84 0 0 0 5.324 7.44c.387.079.528-.168.528-.376 0-.182-.013-.805-.013-1.454-2.165.467-2.616-.935-2.616-.935-.349-.91-.864-1.143-.864-1.143-.71-.48.051-.48.051-.48.787.051 1.2.805 1.2.805.695 1.194 1.817.857 2.268.649.064-.507.27-.857.49-1.052-1.728-.182-3.545-.857-3.545-3.87 0-.857.31-1.558.8-2.104-.078-.195-.349-1 .077-2.078 0 0 .657-.208 2.14.805a7.5 7.5 0 0 1 1.946-.26c.657 0 1.328.092 1.946.26 1.483-1.013 2.14-.805 2.14-.805.426 1.078.155 1.883.078 2.078.502.546.799 1.247.799 2.104 0 3.013-1.818 3.675-3.558 3.87.284.247.528.714.528 1.454 0 1.052-.012 1.896-.012 2.156 0 .208.142.455.528.377a7.84 7.84 0 0 0 5.324-7.441c.013-4.338-3.48-7.844-7.773-7.844" clip-rule="evenodd"/>
+  </symbol>
+  <symbol id="social-icon" viewBox="0 0 20 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M12.5 6.667a4.167 4.167 0 1 0-8.334 0 4.167 4.167 0 0 0 8.334 0"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M2.5 16.667a5.833 5.833 0 0 1 8.75-5.053m3.837.474.513 1.035c.07.144.257.282.414.309l.93.155c.596.1.736.536.307.965l-.723.73a.64.64 0 0 0-.152.531l.207.903c.164.715-.213.991-.84.618l-.872-.52a.63.63 0 0 0-.577 0l-.872.52c-.624.373-1.003.094-.84-.618l.207-.903a.64.64 0 0 0-.152-.532l-.723-.729c-.426-.43-.289-.864.306-.964l.93-.156a.64.64 0 0 0 .412-.31l.513-1.034c.28-.562.735-.562 1.012 0"/>
+  </symbol>
+  <symbol id="x-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M1.893 1.98c.052.072 1.245 1.769 2.653 3.77l2.892 4.114c.183.261.333.48.333.486s-.068.089-.152.183l-.522.593-.765.867-3.597 4.087c-.375.426-.734.834-.798.905a1 1 0 0 0-.118.148c0 .01.236.017.664.017h.663l.729-.83c.4-.457.796-.906.879-.999a692 692 0 0 0 1.794-2.038c.034-.037.301-.34.594-.675l.551-.624.345-.392a7 7 0 0 1 .34-.374c.006 0 .93 1.306 2.052 2.903l2.084 2.965.045.063h2.275c1.87 0 2.273-.003 2.266-.021-.008-.02-1.098-1.572-3.894-5.547-2.013-2.862-2.28-3.246-2.273-3.266.008-.019.282-.332 2.085-2.38l2-2.274 1.567-1.782c.022-.028-.016-.03-.65-.03h-.674l-.3.342a871 871 0 0 1-1.782 2.025c-.067.075-.405.458-.75.852a100 100 0 0 1-.803.91c-.148.172-.299.344-.99 1.127-.304.343-.32.358-.345.327-.015-.019-.904-1.282-1.976-2.808L6.365 1.85H1.8zm1.782.91 8.078 11.294c.772 1.08 1.413 1.973 1.425 1.984.016.017.241.02 1.05.017l1.03-.004-2.694-3.766L7.796 5.75 5.722 2.852l-1.039-.004-1.039-.004z" clip-rule="evenodd"/>
+  </symbol>
+</svg>

frontend/src/App.vue

@@ -0,0 +1,78 @@
+<template>
+  <div id="app-root">
+    <AppHeader @open-settings="showSettings = true" />
+    <AppStepper v-if="currentStep > 0" :current-step="currentStep" />
+    <main class="main-area">
+      <router-view v-slot="{ Component }">
+        <transition name="fade-slide" mode="out-in">
+          <component :is="Component" ref="currentViewRef" @next="handleNext" @restart="handleRestart" />
+        </transition>
+      </router-view>
+    </main>
+    <BottomNav v-if="currentStep > 0" :current-step="currentStep" :hint="stepHints[currentStep]"
+      :disabled="isExtracting" @next="handleNext" />
+    <SettingsModal v-if="showSettings" @close="showSettings = false" />
+    <Toast />
+  </div>
+</template>
+
+<script setup>
+import { ref, computed, provide } from 'vue'
+import { useRouter, useRoute } from 'vue-router'
+import AppHeader from './components/AppHeader.vue'
+import AppStepper from './components/AppStepper.vue'
+import BottomNav from './components/BottomNav.vue'
+import SettingsModal from './components/SettingsModal.vue'
+import Toast from './components/Toast.vue'
+
+const router = useRouter()
+const route = useRoute()
+const showSettings = ref(false)
+const currentViewRef = ref(null)
+const isExtracting = ref(false)
+
+provide('setExtracting', (val) => { isExtracting.value = val })
+
+const stepMap = { '/': 0, '/upload': 1, '/extract': 2, '/match': 3, '/export': 4 }
+const currentStep = computed(() => stepMap[route.path] ?? 0)
+
+const stepHints = {
+  0: '欢迎使用故事提取工具',
+  1: '上传文字稿和讲述者信息文件后进入下一步',
+  2: 'AI 将自动从文字稿中提取个人故事',
+  3: '将提取的故事与讲述者一一对应',
+  4: '预览最终文档效果并导出',
+}
+
+function handleNext() {
+  if (isExtracting.value) return
+  if (currentStep.value === 4) {
+    if (currentViewRef.value && currentViewRef.value.handleDownloadAll) {
+      currentViewRef.value.handleDownloadAll()
+    }
+  } else if (currentStep.value < 4) {
+    router.push(['/', '/upload', '/extract', '/match', '/export'][currentStep.value + 1])
+  }
+}
+
+function handleRestart() {
+  if (isExtracting.value) return
+  localStorage.removeItem('extraction_task_id')
+  router.push('/')
+}
+</script>
+
+<style>
+.fade-slide-enter-active,
+.fade-slide-leave-active {
+  transition: all 0.3s ease;
+}
+.fade-slide-enter-from {
+  opacity: 0;
+  transform: translateY(12px);
+}
+.fade-slide-leave-to {
+  opacity: 0;
+  transform: translateY(-12px);
+}
+</style>

frontend/src/api/index.js

@@ -0,0 +1,60 @@
+import axios from 'axios'
+
+const api = axios.create({
+  baseURL: '/api/v1',
+  timeout: 300000,
+})
+
+// ===== Files =====
+export const uploadTranscripts = (files) => {
+  const formData = new FormData()
+  files.forEach(f => formData.append('files', f))
+  return api.post('/files/transcripts', formData, { headers: { 'Content-Type': 'multipart/form-data' } })
+}
+
+export const uploadPersons = (files) => {
+  const formData = new FormData()
+  files.forEach(f => formData.append('files', f))
+  return api.post('/files/persons', formData, { headers: { 'Content-Type': 'multipart/form-data' } })
+}
+
+export const getTranscripts = () => api.get('/files/transcripts')
+export const getPersons = () => api.get('/files/persons')
+export const deleteTranscript = (id) => api.delete(`/files/transcripts/${id}`)
+export const deletePerson = (id) => api.delete(`/files/persons/${id}`)
+export const previewPerson = (id) => api.get(`/files/persons/${id}/preview`)
+
+// ===== Extraction =====
+export const startExtraction = (taskId) => api.post('/extraction/start', null, { params: taskId ? { task_id: taskId } : {} })
+export const getExtractionStatus = (taskId) => api.get('/extraction/status', { params: taskId ? { task_id: taskId } : {} })
+export const streamExtractionStatus = (taskId) => {
+  const url = taskId
+    ? `/api/v1/extraction/status/stream?task_id=${taskId}`
+    : '/api/v1/extraction/status/stream'
+  return new EventSource(url)
+}
+export const stopExtraction = (taskId) => api.post('/extraction/stop', null, { params: taskId ? { task_id: taskId } : {} })
+
+// ===== Stories =====
+export const getStories = (status = 'all') => api.get('/stories', { params: { status } })
+export const getStory = (id) => api.get(`/stories/${id}`)
+export const editStory = (id, data) => api.put(`/stories/${id}`, data)
+export const deleteStory = (id) => api.delete(`/stories/${id}`)
+
+// ===== Matching =====
+export const createMatch = (storyId, personId) => api.post('/matches', { story_id: storyId, person_id: personId })
+export const removeMatch = (storyId) => api.delete(`/matches/${storyId}`)
+export const getMatches = () => api.get('/matches')
+
+// ===== Export =====
+export const getExportList = () => api.get('/export/list')
+export const previewExport = (personId) => api.get(`/export/preview/${personId}`)
+export const downloadOne = (personId) => api.get(`/export/download/${personId}`, { responseType: 'blob' })
+export const downloadAll = () => api.get('/export/download-all', { responseType: 'blob' })
+
+// ===== Settings =====
+export const getSettings = () => api.get('/settings')
+export const updateSettings = (data) => api.put('/settings', data)
+export const testLlm = () => api.post('/settings/test-llm')
+
+export default api

frontend/src/assets/styles/global.css

@@ -0,0 +1,936 @@
+* { font-family: 'Inter', -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif; box-sizing: border-box; }
+
+:root {
+    --primary: #6366f1;
+    --primary-light: #818cf8;
+    --primary-dark: #4f46e5;
+    --primary-bg: #eef2ff;
+    --accent: #f59e0b;
+    --success: #10b981;
+    --success-bg: #ecfdf5;
+    --danger: #ef4444;
+    --bg: #f1f5f9;
+    --card: #ffffff;
+    --border: #e2e8f0;
+    --border-light: #f1f5f9;
+    --text: #0f172a;
+    --text-secondary: #475569;
+    --text-muted: #94a3b8;
+}
+
+body {
+    background: var(--bg);
+    color: var(--text);
+    margin: 0;
+    min-height: 100vh;
+}
+
+/* ===== HEADER ===== */
+.app-header {
+    background: white;
+    border-bottom: 1px solid var(--border);
+    padding: 0 40px;
+    height: 64px;
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    position: sticky;
+    top: 0;
+    z-index: 100;
+}
+
+.app-logo {
+    display: flex;
+    align-items: center;
+    gap: 12px;
+}
+
+.app-logo-icon {
+    width: 36px;
+    height: 36px;
+    border-radius: 10px;
+    background: linear-gradient(135deg, var(--primary), var(--primary-dark));
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    color: white;
+    font-size: 16px;
+}
+
+.app-logo-text {
+    font-size: 17px;
+    font-weight: 700;
+    color: var(--text);
+}
+
+.app-logo-text span {
+    color: var(--text-muted);
+    font-weight: 400;
+    font-size: 13px;
+    margin-left: 8px;
+}
+
+.header-actions {
+    display: flex;
+    align-items: center;
+    gap: 8px;
+}
+
+.header-btn {
+    width: 36px;
+    height: 36px;
+    border-radius: 10px;
+    border: 1px solid var(--border);
+    background: white;
+    cursor: pointer;
+    color: var(--text-secondary);
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 14px;
+    transition: all 0.2s;
+}
+
+.header-btn:hover {
+    border-color: var(--primary-light);
+    color: var(--primary);
+    background: var(--primary-bg);
+}
+
+/* ===== STEPPER ===== */
+.stepper-wrapper {
+    background: white;
+    border-bottom: 1px solid var(--border);
+    padding: 0 40px;
+}
+
+.stepper {
+    display: flex;
+    align-items: center;
+    max-width: 800px;
+    margin: 0 auto;
+    padding: 20px 0;
+}
+
+.stepper-step {
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+    flex: 1;
+    position: relative;
+    cursor: pointer;
+}
+
+.stepper-step:not(:last-child)::after {
+    content: '';
+    position: absolute;
+    top: 18px;
+    left: calc(50% + 24px);
+    width: calc(100% - 48px);
+    height: 3px;
+    background: var(--border);
+    border-radius: 2px;
+    transition: background 0.4s;
+}
+
+.stepper-step.completed:not(:last-child)::after {
+    background: var(--success);
+}
+
+.stepper-step.active:not(:last-child)::after {
+    background: linear-gradient(90deg, var(--success) 50%, var(--border) 50%);
+}
+
+.step-circle {
+    width: 36px;
+    height: 36px;
+    border-radius: 50%;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 14px;
+    font-weight: 700;
+    background: var(--border-light);
+    color: var(--text-muted);
+    border: 3px solid white;
+    box-shadow: 0 0 0 1px var(--border);
+    transition: all 0.3s;
+    position: relative;
+    z-index: 2;
+}
+
+.stepper-step.active .step-circle {
+    background: var(--primary);
+    color: white;
+    box-shadow: 0 0 0 3px var(--primary-bg), 0 4px 12px rgba(99,102,241,0.25);
+}
+
+.stepper-step.completed .step-circle {
+    background: var(--success);
+    color: white;
+    box-shadow: 0 0 0 3px var(--success-bg);
+}
+
+.step-label {
+    margin-top: 8px;
+    font-size: 13px;
+    font-weight: 600;
+    color: var(--text-muted);
+    transition: color 0.3s;
+}
+
+.stepper-step.active .step-label {
+    color: var(--primary);
+}
+
+.stepper-step.completed .step-label {
+    color: var(--success);
+}
+
+/* ===== MAIN CONTENT ===== */
+.main-area {
+    max-width: 1100px;
+    margin: 0 auto;
+    padding: 28px 40px 120px;
+}
+
+.step-page { display: none; }
+.step-page.active { display: block; animation: fadeSlideIn 0.35s ease; }
+
+@keyframes fadeSlideIn {
+    from { opacity: 0; transform: translateY(12px); }
+    to { opacity: 1; transform: translateY(0); }
+}
+
+/* ===== BOTTOM NAV ===== */
+.bottom-nav {
+    position: fixed;
+    bottom: 0;
+    left: 0;
+    right: 0;
+    background: white;
+    border-top: 1px solid var(--border);
+    padding: 14px 40px;
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    z-index: 100;
+}
+
+.bottom-nav-hint {
+    font-size: 13px;
+    color: var(--text-muted);
+}
+
+.bottom-nav-actions {
+    display: flex;
+    gap: 10px;
+}
+
+/* ===== BUTTONS ===== */
+.btn {
+    display: inline-flex;
+    align-items: center;
+    gap: 8px;
+    padding: 10px 22px;
+    border-radius: 10px;
+    font-size: 14px;
+    font-weight: 600;
+    cursor: pointer;
+    transition: all 0.2s;
+    border: none;
+    white-space: nowrap;
+}
+
+.btn-primary {
+    background: var(--primary);
+    color: white;
+}
+.btn-primary:hover {
+    background: var(--primary-dark);
+    box-shadow: 0 4px 14px rgba(99,102,241,0.3);
+}
+
+.btn-success {
+    background: var(--success);
+    color: white;
+}
+.btn-success:hover { background: #059669; }
+
+.btn-outline {
+    background: white;
+    color: var(--text-secondary);
+    border: 1px solid var(--border);
+}
+.btn-outline:hover {
+    border-color: var(--primary-light);
+    color: var(--primary);
+    background: var(--primary-bg);
+}
+
+.btn-ghost {
+    background: transparent;
+    color: var(--text-secondary);
+    padding: 10px 16px;
+}
+.btn-ghost:hover { color: var(--primary); background: var(--primary-bg); }
+
+.btn-sm { padding: 7px 14px; font-size: 13px; border-radius: 8px; }
+.btn-lg { padding: 12px 28px; font-size: 15px; border-radius: 12px; }
+
+/* ===== CARDS ===== */
+.card {
+    background: var(--card);
+    border: 1px solid var(--border);
+    border-radius: 16px;
+    padding: 24px;
+    margin-bottom: 20px;
+}
+
+.card-header {
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+    margin-bottom: 20px;
+}
+
+.card-title {
+    font-size: 16px;
+    font-weight: 700;
+    color: var(--text);
+    display: flex;
+    align-items: center;
+    gap: 10px;
+}
+
+.card-title i {
+    width: 32px;
+    height: 32px;
+    border-radius: 8px;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 14px;
+}
+
+.card-title .icon-upload { background: #dbeafe; color: #2563eb; }
+.card-title .icon-person { background: #fce7f3; color: #db2777; }
+.card-title .icon-magic { background: #ede9fe; color: #7c3aed; }
+.card-title .icon-link { background: #fef3c7; color: #d97706; }
+.card-title .icon-export { background: #d1fae5; color: #059669; }
+
+.card-badge {
+    font-size: 12px;
+    color: var(--text-muted);
+    font-weight: 500;
+}
+
+/* ===== UPLOAD ZONE ===== */
+.upload-zone {
+    border: 2px dashed var(--border);
+    border-radius: 14px;
+    padding: 36px;
+    text-align: center;
+    cursor: pointer;
+    transition: all 0.3s;
+    background: #fafbfd;
+}
+.upload-zone:hover {
+    border-color: var(--primary-light);
+    background: var(--primary-bg);
+}
+.upload-zone i.upload-icon {
+    font-size: 32px;
+    color: var(--text-muted);
+    margin-bottom: 10px;
+    display: block;
+}
+.upload-zone p { color: var(--text-secondary); font-size: 14px; margin: 3px 0; }
+.upload-zone .hint { color: var(--text-muted); font-size: 12px; }
+
+/* ===== FILE LIST ===== */
+.file-list { margin-top: 14px; }
+
+.file-item {
+    display: flex;
+    align-items: center;
+    gap: 14px;
+    padding: 12px 14px;
+    border: 1px solid var(--border-light);
+    border-radius: 10px;
+    margin-bottom: 6px;
+    transition: all 0.2s;
+}
+.file-item:hover {
+    background: #fafbfd;
+    border-color: var(--border);
+}
+
+.file-icon {
+    width: 40px;
+    height: 40px;
+    border-radius: 10px;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 16px;
+    flex-shrink: 0;
+}
+.file-icon.docx { background: #dbeafe; color: #2563eb; }
+.file-icon.person { background: #fce7f3; color: #db2777; }
+
+.file-info { flex: 1; min-width: 0; }
+.file-name { font-size: 14px; font-weight: 600; color: var(--text); white-space: nowrap; overflow: hidden; text-overflow: ellipsis; }
+.file-meta { font-size: 12px; color: var(--text-muted); margin-top: 2px; }
+
+.file-action-btn {
+    width: 30px; height: 30px; border-radius: 8px; border: 1px solid transparent;
+    background: transparent; cursor: pointer; color: var(--text-muted);
+    display: flex; align-items: center; justify-content: center; font-size: 13px;
+    transition: all 0.2s;
+}
+.file-action-btn:hover { border-color: var(--border); background: white; color: var(--text-secondary); }
+.file-action-btn.danger:hover { border-color: #fecaca; background: #fef2f2; color: var(--danger); }
+
+/* ===== STATS ROW ===== */
+.stats-row {
+    display: grid;
+    grid-template-columns: repeat(4, 1fr);
+    gap: 14px;
+    margin-bottom: 24px;
+}
+
+.stat-item {
+    background: white;
+    border: 1px solid var(--border);
+    border-radius: 12px;
+    padding: 16px 18px;
+    display: flex;
+    align-items: center;
+    gap: 14px;
+}
+
+.stat-icon {
+    width: 42px;
+    height: 42px;
+    border-radius: 10px;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 16px;
+    flex-shrink: 0;
+}
+.stat-icon.blue { background: #eef2ff; color: var(--primary); }
+.stat-icon.amber { background: #fffbeb; color: var(--accent); }
+.stat-icon.green { background: #ecfdf5; color: var(--success); }
+.stat-icon.rose { background: #fff1f2; color: #e11d48; }
+
+.stat-value { font-size: 22px; font-weight: 800; color: var(--text); line-height: 1; }
+.stat-label { font-size: 12px; color: var(--text-muted); margin-top: 2px; }
+
+/* ===== TWO COL ===== */
+.two-col {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 20px;
+}
+
+/* ===== TAG CHIPS ===== */
+.chip {
+    display: inline-flex;
+    align-items: center;
+    gap: 6px;
+    padding: 6px 14px;
+    border-radius: 8px;
+    font-size: 13px;
+    font-weight: 500;
+    cursor: pointer;
+    transition: all 0.2s;
+}
+.chip.selected { background: var(--primary-bg); border: 1px solid #c7d2fe; color: var(--primary); }
+.chip.unselected { background: white; border: 1px solid var(--border); color: var(--text-secondary); }
+.chip.unselected:hover { border-color: var(--primary-light); color: var(--primary); }
+
+/* ===== PROGRESS ===== */
+.progress-bar {
+    height: 8px;
+    background: var(--border-light);
+    border-radius: 4px;
+    overflow: hidden;
+}
+.progress-fill {
+    height: 100%;
+    border-radius: 4px;
+    background: linear-gradient(90deg, var(--primary), var(--primary-light));
+    transition: width 0.5s ease;
+}
+
+.extraction-status {
+    display: flex;
+    align-items: center;
+    gap: 12px;
+    padding: 14px 18px;
+    background: #eff6ff;
+    border: 1px solid #bfdbfe;
+    border-radius: 12px;
+}
+
+.status-dot {
+    width: 10px;
+    height: 10px;
+    border-radius: 50%;
+    flex-shrink: 0;
+}
+.status-dot.running { background: var(--primary); animation: pulse 1.5s infinite; }
+.status-dot.done { background: var(--success); }
+
+@keyframes pulse {
+    0%, 100% { opacity: 1; }
+    50% { opacity: 0.3; }
+}
+
+/* ===== STORY CARDS ===== */
+.story-card {
+    background: white;
+    border: 1px solid var(--border);
+    border-radius: 14px;
+    padding: 18px 20px;
+    margin-bottom: 10px;
+    transition: all 0.2s;
+}
+.story-card:hover {
+    border-color: #cbd5e1;
+    box-shadow: 0 2px 12px rgba(0,0,0,0.04);
+}
+
+.story-title {
+    font-size: 15px;
+    font-weight: 600;
+    color: var(--text);
+}
+
+.story-tag {
+    display: inline-flex;
+    align-items: center;
+    gap: 4px;
+    padding: 3px 10px;
+    border-radius: 6px;
+    font-size: 11px;
+    font-weight: 600;
+    margin-top: 6px;
+}
+.story-tag.pending { background: #fef3c7; color: #92400e; }
+.story-tag.matched { background: #d1fae5; color: #065f46; }
+
+.story-summary {
+    font-size: 13px;
+    color: var(--text-secondary);
+    line-height: 1.6;
+    margin-top: 8px;
+}
+
+.story-source {
+    font-size: 12px;
+    color: var(--text-muted);
+    margin-top: 8px;
+    display: flex;
+    align-items: center;
+    gap: 4px;
+}
+
+.story-actions {
+    display: flex;
+    gap: 6px;
+    margin-top: 10px;
+}
+
+/* ===== TABS ===== */
+.tabs {
+    display: flex;
+    gap: 4px;
+    background: var(--border-light);
+    padding: 4px;
+    border-radius: 10px;
+    margin-bottom: 16px;
+}
+.tab {
+    padding: 7px 16px;
+    border-radius: 8px;
+    font-size: 13px;
+    font-weight: 600;
+    cursor: pointer;
+    color: var(--text-muted);
+    transition: all 0.2s;
+    border: none;
+    background: transparent;
+}
+.tab.active { background: white; color: var(--primary); box-shadow: 0 1px 3px rgba(0,0,0,0.06); }
+.tab:hover:not(.active) { color: var(--text-secondary); }
+
+/* ===== MATCH PANEL ===== */
+.match-panel {
+    display: grid;
+    grid-template-columns: 1fr 60px 1fr;
+    gap: 0;
+    align-items: start;
+}
+
+.match-center-col {
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+    padding-top: 60px;
+}
+
+.match-arrow-btn {
+    width: 44px;
+    height: 44px;
+    border-radius: 50%;
+    background: var(--primary);
+    color: white;
+    border: none;
+    cursor: pointer;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 16px;
+    box-shadow: 0 4px 14px rgba(99,102,241,0.3);
+    transition: all 0.2s;
+}
+.match-arrow-btn:hover {
+    transform: scale(1.1);
+    box-shadow: 0 6px 20px rgba(99,102,241,0.4);
+}
+
+.match-hint {
+    font-size: 11px;
+    color: var(--text-muted);
+    margin-top: 8px;
+    text-align: center;
+    line-height: 1.4;
+}
+
+.person-card {
+    display: flex;
+    align-items: center;
+    gap: 12px;
+    padding: 12px 14px;
+    border: 1px solid var(--border-light);
+    border-radius: 10px;
+    margin-bottom: 6px;
+    cursor: pointer;
+    transition: all 0.2s;
+}
+.person-card:hover { border-color: var(--border); background: #fafbfd; }
+.person-card.selected { border-color: var(--primary); background: var(--primary-bg); }
+
+.person-avatar {
+    width: 42px;
+    height: 42px;
+    border-radius: 10px;
+    background: linear-gradient(135deg, #e0e7ff, #c7d2fe);
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    color: var(--primary);
+    font-size: 16px;
+    font-weight: 700;
+    flex-shrink: 0;
+}
+
+.person-name { font-size: 14px; font-weight: 600; color: var(--text); }
+.person-meta { font-size: 12px; color: var(--text-muted); margin-top: 1px; }
+
+/* ===== EXPORT ===== */
+.export-item {
+    display: flex;
+    align-items: center;
+    gap: 14px;
+    padding: 14px 16px;
+    border: 1px solid var(--border-light);
+    border-radius: 12px;
+    margin-bottom: 8px;
+    transition: all 0.2s;
+}
+.export-item:hover { border-color: var(--border); background: #fafbfd; }
+
+.export-check {
+    width: 26px;
+    height: 26px;
+    border-radius: 50%;
+    background: #d1fae5;
+    color: var(--success);
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 12px;
+    flex-shrink: 0;
+}
+.export-check.pending { background: var(--border-light); color: var(--text-muted); }
+
+/* ===== PREVIEW DOC ===== */
+.preview-doc {
+    background: white;
+    border: 1px solid var(--border);
+    border-radius: 14px;
+    overflow: hidden;
+    box-shadow: 0 8px 30px rgba(0,0,0,0.06);
+}
+
+.preview-doc-header {
+    background: linear-gradient(135deg, var(--primary), var(--primary-dark));
+    padding: 28px 24px;
+    color: white;
+    text-align: center;
+}
+
+.preview-doc-avatar {
+    width: 68px;
+    height: 68px;
+    border-radius: 50%;
+    border: 3px solid rgba(255,255,255,0.3);
+    margin: 0 auto 12px;
+    background: rgba(255,255,255,0.2);
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 26px;
+}
+
+.preview-doc-name { font-size: 18px; font-weight: 700; }
+.preview-doc-info { font-size: 13px; opacity: 0.8; margin-top: 4px; }
+
+.preview-doc-body { padding: 24px; }
+.preview-doc-body h3 {
+    font-size: 15px;
+    font-weight: 700;
+    color: var(--text);
+    margin-bottom: 12px;
+    padding-bottom: 8px;
+    border-bottom: 2px solid var(--primary);
+    display: inline-block;
+}
+.preview-doc-body p {
+    font-size: 13px;
+    color: var(--text-secondary);
+    line-height: 1.8;
+}
+
+/* ===== SETTINGS MODAL ===== */
+.modal-overlay {
+    display: none;
+    position: fixed;
+    inset: 0;
+    background: rgba(0,0,0,0.4);
+    z-index: 200;
+    align-items: center;
+    justify-content: center;
+}
+.modal-overlay.open { display: flex; }
+
+.modal {
+    background: white;
+    border-radius: 18px;
+    width: 560px;
+    max-height: 80vh;
+    overflow-y: auto;
+    box-shadow: 0 20px 60px rgba(0,0,0,0.15);
+    animation: fadeSlideIn 0.25s ease;
+}
+
+.modal-header {
+    padding: 20px 24px;
+    border-bottom: 1px solid var(--border);
+    display: flex;
+    align-items: center;
+    justify-content: space-between;
+}
+
+.modal-header h2 { font-size: 17px; font-weight: 700; margin: 0; }
+
+.modal-close {
+    width: 32px; height: 32px; border-radius: 8px; border: none;
+    background: var(--border-light); cursor: pointer; color: var(--text-muted);
+    display: flex; align-items: center; justify-content: center; font-size: 14px;
+    transition: all 0.2s;
+}
+.modal-close:hover { background: #fee2e2; color: var(--danger); }
+
+.modal-body { padding: 24px; }
+
+.form-group { margin-bottom: 16px; }
+.form-label {
+    font-size: 13px;
+    font-weight: 600;
+    color: var(--text);
+    display: block;
+    margin-bottom: 6px;
+}
+.form-input {
+    width: 100%;
+    padding: 9px 14px;
+    border: 1px solid var(--border);
+    border-radius: 10px;
+    font-size: 14px;
+    color: var(--text);
+    transition: border-color 0.2s;
+}
+.form-input:focus { outline: none; border-color: var(--primary); box-shadow: 0 0 0 3px var(--primary-bg); }
+
+.form-row {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 14px;
+}
+
+.modal-footer {
+    padding: 16px 24px;
+    border-top: 1px solid var(--border);
+    display: flex;
+    justify-content: flex-end;
+    gap: 8px;
+}
+
+/* ===== SCROLLBAR ===== */
+::-webkit-scrollbar { width: 5px; }
+::-webkit-scrollbar-track { background: transparent; }
+::-webkit-scrollbar-thumb { background: #cbd5e1; border-radius: 3px; }
+::-webkit-scrollbar-thumb:hover { background: #94a3b8; }
+
+/* ===== MISC ===== */
+.text-muted { color: var(--text-muted); }
+.text-sm { font-size: 13px; }
+.mt-2 { margin-top: 8px; }
+.mt-4 { margin-top: 16px; }
+.mb-4 { margin-bottom: 16px; }
+.flex { display: flex; }
+.items-center { align-items: center; }
+.gap-2 { gap: 8px; }
+.gap-3 { gap: 12px; }
+.flex-1 { flex: 1; }
+.text-right { text-align: right; }
+
+.match-highlight {
+    border-color: var(--accent) !important;
+    background: #fffbeb !important;
+}
+
+.scrollable { max-height: 65vh; overflow-y: auto; }
+
+/* ===== WELCOME PAGE ===== */
+.welcome-page {
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+    justify-content: center;
+    min-height: calc(100vh - 240px);
+    text-align: center;
+    padding: 40px 20px;
+}
+
+.welcome-icon {
+    width: 88px;
+    height: 88px;
+    border-radius: 24px;
+    background: linear-gradient(135deg, var(--primary), #8b5cf6);
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 36px;
+    color: white;
+    margin-bottom: 28px;
+    box-shadow: 0 12px 40px rgba(99,102,241,0.3);
+}
+
+.welcome-title {
+    font-size: 32px;
+    font-weight: 800;
+    color: var(--text);
+    margin-bottom: 12px;
+    letter-spacing: -0.5px;
+}
+
+.welcome-desc {
+    font-size: 16px;
+    color: var(--text-secondary);
+    line-height: 1.7;
+    max-width: 480px;
+    margin-bottom: 36px;
+}
+
+.welcome-steps {
+    display: flex;
+    gap: 32px;
+    margin-bottom: 40px;
+}
+
+.welcome-step-item {
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+    gap: 10px;
+    width: 120px;
+}
+
+.welcome-step-icon {
+    width: 48px;
+    height: 48px;
+    border-radius: 14px;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    font-size: 18px;
+}
+
+.welcome-step-icon.s1 { background: #dbeafe; color: #2563eb; }
+.welcome-step-icon.s2 { background: #ede9fe; color: #7c3aed; }
+.welcome-step-icon.s3 { background: #fef3c7; color: #d97706; }
+.welcome-step-icon.s4 { background: #d1fae5; color: #059669; }
+
+.welcome-step-label {
+    font-size: 13px;
+    font-weight: 600;
+    color: var(--text-secondary);
+}
+
+.welcome-step-num {
+    font-size: 11px;
+    color: var(--text-muted);
+}
+
+/* ===== TOAST ===== */
+.toast-container {
+    position: fixed;
+    top: 80px;
+    left: 50%;
+    transform: translateX(-50%);
+    z-index: 300;
+    pointer-events: none;
+}
+
+.toast {
+    background: #1e293b;
+    color: white;
+    padding: 12px 20px;
+    border-radius: 10px;
+    font-size: 14px;
+    font-weight: 500;
+    display: flex;
+    align-items: center;
+    gap: 10px;
+    box-shadow: 0 8px 30px rgba(0,0,0,0.2);
+    animation: toastIn 0.3s ease, toastOut 0.3s ease 2.5s forwards;
+    pointer-events: auto;
+}
+
+.toast i { color: var(--accent); font-size: 16px; }
+
+@keyframes toastIn {
+    from { opacity: 0; transform: translateY(-12px); }
+    to { opacity: 1; transform: translateY(0); }
+}
+
+@keyframes toastOut {
+    from { opacity: 1; transform: translateY(0); }
+    to { opacity: 0; transform: translateY(-12px); }
+}

frontend/src/assets/vue.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" class="iconify iconify--logos" width="37.07" height="36" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 198"><path fill="#41B883" d="M204.8 0H256L128 220.8L0 0h97.92L128 51.2L157.44 0h47.36Z"></path><path fill="#41B883" d="m0 0l128 220.8L256 0h-51.2L128 132.48L50.56 0H0Z"></path><path fill="#35495E" d="M50.56 0L128 133.12L204.8 0h-47.36L128 51.2L97.92 0H50.56Z"></path></svg>

frontend/src/components/AppHeader.vue

@@ -0,0 +1,22 @@
+<template>
+  <header class="app-header">
+    <div class="app-logo">
+      <div class="app-logo-icon">
+        <i class="fas fa-book-open"></i>
+      </div>
+      <div class="app-logo-text">
+        Story Extractor
+        <span>直播文字稿故事提取工具</span>
+      </div>
+    </div>
+    <div class="header-actions">
+      <button class="header-btn" title="设置" @click="$emit('open-settings')">
+        <i class="fas fa-gear"></i>
+      </button>
+    </div>
+  </header>
+</template>
+
+<script setup>
+defineEmits(['open-settings'])
+</script>

frontend/src/components/AppStepper.vue

@@ -0,0 +1,30 @@
+<template>
+  <div class="stepper-wrapper">
+    <div class="stepper">
+      <div
+        v-for="(step, index) in steps"
+        :key="index"
+        class="stepper-step"
+        :class="{ active: currentStep === index + 1, completed: currentStep > index + 1 }"
+      >
+        <div class="step-circle">
+          <template v-if="currentStep > index + 1">
+            <i class="fas fa-check"></i>
+          </template>
+          <template v-else>
+            {{ index + 1 }}
+          </template>
+        </div>
+        <span class="step-label">{{ step }}</span>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup>
+defineProps({
+  currentStep: { type: Number, required: true }
+})
+
+const steps = ['上传文件', '提取故事', '匹配确认', '预览导出']
+</script>

frontend/src/components/BottomNav.vue

@@ -0,0 +1,37 @@
+<template>
+  <div class="bottom-nav">
+    <div class="bottom-nav-hint">
+      <span>{{ hint }}</span>
+    </div>
+    <div class="bottom-nav-actions">
+      <button
+        v-if="currentStep === 4"
+        class="btn btn-success"
+        :disabled="disabled"
+        @click="$emit('next')"
+      >
+        <i class="fas fa-download"></i>
+        全部导出
+      </button>
+      <button
+        v-else
+        class="btn btn-primary"
+        :disabled="disabled"
+        @click="$emit('next')"
+      >
+        下一步
+        <i class="fas fa-chevron-right"></i>
+      </button>
+    </div>
+  </div>
+</template>
+
+<script setup>
+defineProps({
+  currentStep: { type: Number, required: true },
+  hint: { type: String, default: '' },
+  disabled: { type: Boolean, default: false },
+})
+
+defineEmits(['next'])
+</script>

frontend/src/components/SettingsModal.vue

@@ -0,0 +1,191 @@
+<template>
+  <div class="modal-overlay open" @click.self="$emit('close')">
+    <div class="modal">
+      <div class="modal-header">
+        <h2><i class="fas fa-cog" style="color:var(--primary); margin-right:8px;"></i>系统设置</h2>
+        <button class="modal-close" @click="$emit('close')">
+          <i class="fas fa-times"></i>
+        </button>
+      </div>
+      <div class="modal-body">
+        <!-- LLM 配置 section -->
+        <div class="settings-section">
+          <div class="settings-section-header">
+            <i class="fas fa-robot"></i>
+            <span>LLM 配置</span>
+          </div>
+          <div class="form-row">
+            <div class="form-group">
+              <label class="form-label">API 提供商</label>
+              <select v-model="form.llm_provider" class="form-input">
+                <option value="openai">OpenAI (GPT-4)</option>
+                <option value="deepseek">DeepSeek</option>
+                <option value="tongyi">通义千问</option>
+                <option value="minimax">MiniMax</option>
+                <option value="custom">自定义兼容 API</option>
+              </select>
+            </div>
+            <div class="form-group">
+              <label class="form-label">模型名称</label>
+              <input v-model="form.llm_model" type="text" class="form-input" placeholder="gpt-4o" />
+            </div>
+          </div>
+          <div class="form-row">
+            <div class="form-group">
+              <label class="form-label">API Base URL</label>
+              <input v-model="form.llm_base_url" type="text" class="form-input" placeholder="https://api.openai.com/v1" />
+            </div>
+            <div class="form-group">
+              <label class="form-label">API Key</label>
+              <input v-model="form.llm_api_key" type="password" class="form-input" placeholder="输入 API Key" />
+            </div>
+          </div>
+        </div>
+
+        <!-- 分隔线 -->
+        <div class="settings-separator"></div>
+
+        <!-- 提取参数 section -->
+        <div class="settings-section">
+          <div class="settings-section-header">
+            <i class="fas fa-sliders-h"></i>
+            <span>提取参数</span>
+          </div>
+          <div class="form-row">
+            <div class="form-group">
+              <label class="form-label">话题分段最大长度（行数）</label>
+              <input v-model.number="form.segment_max_lines" type="number" class="form-input" placeholder="500" />
+            </div>
+            <div class="form-group">
+              <label class="form-label">故事最小长度（行数）</label>
+              <input v-model.number="form.story_min_lines" type="number" class="form-input" placeholder="50" />
+            </div>
+          </div>
+          <div class="form-row">
+            <div class="form-group">
+              <label class="form-label">置信度阈值</label>
+              <input v-model.number="form.confidence_threshold" type="number" class="form-input" step="0.1" min="0" max="1" placeholder="0.6" />
+            </div>
+            <div class="form-group">
+              <label class="form-label">Temperature</label>
+              <input v-model.number="form.temperature" type="number" class="form-input" step="0.1" min="0" max="1" placeholder="0.3" />
+            </div>
+          </div>
+        </div>
+      </div>
+      <div class="modal-footer">
+        <button class="btn btn-outline" @click="$emit('close')">取消</button>
+        <button class="btn btn-outline" :disabled="testing" @click="handleTest">
+          <i v-if="testing" class="fas fa-spinner fa-spin"></i>
+          <i v-else class="fas fa-plug"></i>
+          {{ testing ? '测试中...' : '测试连接' }}
+        </button>
+        <button class="btn btn-primary" :disabled="saving" @click="handleSave">
+          <i v-if="saving" class="fas fa-spinner fa-spin"></i>
+          <i v-else class="fas fa-save"></i>
+          {{ saving ? '保存中...' : '保存' }}
+        </button>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { ref, onMounted } from 'vue'
+import { getSettings, updateSettings, testLlm } from '../api'
+import { useToast } from '../composables/useToast'
+
+const { showToast } = useToast()
+
+const emit = defineEmits(['close'])
+
+const saving = ref(false)
+const testing = ref(false)
+
+const defaultForm = {
+  llm_provider: 'openai',
+  llm_model: '',
+  llm_base_url: '',
+  llm_api_key: '',
+  segment_max_lines: 500,
+  story_min_lines: 50,
+  confidence_threshold: 0.6,
+  temperature: 0.3,
+}
+
+const form = ref({ ...defaultForm })
+
+onMounted(async () => {
+  try {
+    const { data } = await getSettings()
+    form.value = {
+      llm_provider: data.llm_provider || defaultForm.llm_provider,
+      llm_model: data.llm_model || '',
+      llm_base_url: data.llm_base_url || '',
+      llm_api_key: data.llm_api_key || '',
+      segment_max_lines: data.segment_max_lines ?? defaultForm.segment_max_lines,
+      story_min_lines: data.story_min_lines ?? defaultForm.story_min_lines,
+      confidence_threshold: data.confidence_threshold ?? defaultForm.confidence_threshold,
+      temperature: data.temperature ?? defaultForm.temperature,
+    }
+  } catch (e) {
+    showToast('获取设置失败: ' + (e.response?.data?.detail || e.message), 'error')
+  }
+})
+
+async function handleSave() {
+  saving.value = true
+  try {
+    await updateSettings(form.value)
+    showToast('设置已保存', 'success')
+    emit('close')
+  } catch (e) {
+    showToast('保存失败: ' + (e.response?.data?.detail || e.message), 'error')
+  } finally {
+    saving.value = false
+  }
+}
+
+async function handleTest() {
+  testing.value = true
+  try {
+    const res = await testLlm()
+    if (res.data && res.data.success) {
+      showToast('连接测试成功: ' + (res.data.message || ''), 'success')
+    } else {
+      showToast('连接测试失败: ' + (res.data?.message || '未知错误'), 'error')
+    }
+  } catch (e) {
+    showToast('连接测试失败: ' + (e.response?.data?.detail || e.message), 'error')
+  } finally {
+    testing.value = false
+  }
+}
+</script>
+
+<style scoped>
+.settings-section {
+  margin-bottom: 4px;
+}
+
+.settings-section-header {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  font-size: 14px;
+  font-weight: 700;
+  color: var(--text);
+  margin-bottom: 16px;
+}
+
+.settings-section-header i {
+  color: var(--primary);
+  font-size: 15px;
+}
+
+.settings-separator {
+  height: 1px;
+  background: var(--border);
+  margin: 20px 0;
+}
+</style>

frontend/src/components/Toast.vue

@@ -0,0 +1,30 @@
+<template>
+  <div class="toast-container">
+    <div v-for="t in toasts" :key="t.id" class="toast" :class="'toast-' + t.type">
+      <i :class="iconClass(t.type)"></i>
+      {{ t.message }}
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { useToast } from '../composables/useToast'
+const { toasts } = useToast()
+
+function iconClass(type) {
+  switch (type) {
+    case 'success':
+      return 'fas fa-check-circle'
+    case 'error':
+      return 'fas fa-times-circle'
+    default:
+      return 'fas fa-triangle-exclamation'
+  }
+}
+</script>
+
+<style scoped>
+.toast-warning i { color: var(--accent); }
+.toast-success i { color: var(--success); }
+.toast-error i { color: var(--danger); }
+</style>

frontend/src/composables/useToast.js

@@ -0,0 +1,15 @@
+import { ref } from 'vue'
+
+const toasts = ref([])
+let id = 0
+
+export function useToast() {
+  function showToast(message, type = 'warning') {
+    const t = { id: ++id, message, type }
+    toasts.value.push(t)
+    setTimeout(() => {
+      toasts.value = toasts.value.filter(x => x.id !== t.id)
+    }, 3000)
+  }
+  return { toasts, showToast }
+}

frontend/src/main.js

@@ -0,0 +1,8 @@
+import { createApp } from 'vue'
+import App from './App.vue'
+import router from './router'
+import './assets/styles/global.css'
+
+const app = createApp(App)
+app.use(router)
+app.mount('#app')

frontend/src/router/index.js

@@ -0,0 +1,16 @@
+import { createRouter, createWebHistory } from 'vue-router'
+
+const routes = [
+  { path: '/', name: 'welcome', component: () => import('../views/WelcomeView.vue') },
+  { path: '/upload', name: 'upload', component: () => import('../views/UploadView.vue') },
+  { path: '/extract', name: 'extract', component: () => import('../views/ExtractView.vue') },
+  { path: '/match', name: 'match', component: () => import('../views/MatchView.vue') },
+  { path: '/export', name: 'export', component: () => import('../views/ExportView.vue') },
+]
+
+const router = createRouter({
+  history: createWebHistory(),
+  routes,
+})
+
+export default router

frontend/src/views/ExportView.vue

@@ -0,0 +1,316 @@
+<template>
+  <div class="step-page active">
+    <div class="two-col">
+      <!-- Left column: Export list -->
+      <div class="card">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="fas fa-file-export icon-export"></i>
+            导出列表
+          </div>
+          <span class="card-badge" v-if="exportList.length">
+            共 {{ exportList.length }} 人
+          </span>
+        </div>
+
+        <div v-if="loading" class="text-muted text-sm" style="text-align: center; padding: 32px 0;">
+          <i class="fas fa-spinner fa-spin"></i> 加载中...
+        </div>
+
+        <div v-else-if="!exportList.length" class="text-muted text-sm" style="text-align: center; padding: 32px 0;">
+          暂无导出数据，请先完成故事匹配。
+        </div>
+
+        <div v-else class="scrollable">
+          <div
+            v-for="item in exportList"
+            :key="item.person_id"
+            class="export-item"
+            :class="{ active: selectedPersonId === item.person_id }"
+          >
+            <div class="export-check" :class="{ pending: !item.story_count }">
+              <i :class="item.story_count ? 'fas fa-check' : 'fas fa-minus'"></i>
+            </div>
+            <div class="export-item-info" style="flex: 1; min-width: 0;">
+              <div class="export-item-name">{{ item.person_name }}</div>
+              <div class="export-item-meta">
+                {{ item.story_count }} 个故事
+                <span v-if="item.has_photo" style="color: var(--success);"> · 有照片</span>
+                <span v-else style="color: var(--text-muted);"> · 无照片</span>
+              </div>
+            </div>
+            <div class="export-item-actions" style="display: flex; gap: 6px; flex-shrink: 0;">
+              <button
+                class="file-action-btn"
+                title="预览"
+                :disabled="!item.story_count"
+                @click="handlePreview(item.person_id)"
+              >
+                <i class="fas fa-eye"></i>
+              </button>
+              <button
+                class="file-action-btn"
+                title="下载"
+                :disabled="!item.story_count"
+                @click="handleDownload(item.person_id, item.person_name)"
+              >
+                <i class="fas fa-download"></i>
+              </button>
+            </div>
+          </div>
+        </div>
+      </div>
+
+      <!-- Right column: Preview -->
+      <div class="card" style="margin-bottom: 0;">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="fas fa-eye icon-export"></i>
+            文档预览
+          </div>
+          <div v-if="previewData" style="display: flex; gap: 6px;">
+            <button
+              class="btn btn-ghost btn-sm"
+              :disabled="!hasPrev"
+              @click="navigatePreview(-1)"
+            >
+              <i class="fas fa-chevron-left"></i>
+            </button>
+            <button
+              class="btn btn-ghost btn-sm"
+              :disabled="!hasNext"
+              @click="navigatePreview(1)"
+            >
+              <i class="fas fa-chevron-right"></i>
+            </button>
+          </div>
+        </div>
+
+        <div v-if="previewLoading" class="text-muted text-sm" style="text-align: center; padding: 48px 0;">
+          <i class="fas fa-spinner fa-spin"></i> 加载预览中...
+        </div>
+
+        <div v-else-if="!previewData" class="text-muted text-sm" style="text-align: center; padding: 48px 0;">
+          <i class="fas fa-file-alt" style="font-size: 28px; display: block; margin-bottom: 12px; color: var(--border);"></i>
+          选择一位讲述者以预览其文档
+        </div>
+
+        <div v-else class="preview-doc scrollable">
+          <div class="preview-doc-header">
+            <div class="preview-doc-avatar">
+              <i class="fas fa-user"></i>
+            </div>
+            <div class="preview-doc-name">{{ previewData.person_name }}</div>
+            <div class="preview-doc-info">{{ previewData.person_info || '暂无个人信息' }}</div>
+          </div>
+          <div class="preview-doc-body">
+            <div v-for="(story, idx) in previewData.stories" :key="idx" style="margin-bottom: 24px;">
+              <div style="display: flex; justify-content: space-between; align-items: center; margin-bottom: 8px;">
+                <h3 style="font-size: 16px; font-weight: 700;">{{ story.title }}</h3>
+                <button
+                  v-if="story.raw_material"
+                  class="btn btn-ghost btn-sm"
+                  style="font-size: 12px; color: var(--primary);"
+                  @click="toggleMaterial(idx)"
+                >
+                  <i :class="showMaterialIdx === idx ? 'fas fa-eye-slash' : 'fas fa-database'"></i>
+                  {{ showMaterialIdx === idx ? '收起素材' : '查看素材' }}
+                </button>
+              </div>
+
+              <!-- 素材 JSON 展示 -->
+              <div v-if="showMaterialIdx === idx && story.raw_material" style="margin-bottom: 12px;">
+                <div style="background: #1e1e2e; border-radius: 8px; padding: 12px; max-height: 400px; overflow: auto;">
+                  <pre style="margin: 0; font-size: 12px; line-height: 1.5; color: #cdd6f4; white-space: pre-wrap; word-break: break-all;">{{ formatMaterial(story.raw_material) }}</pre>
+                </div>
+              </div>
+
+              <!-- 故事梗概 -->
+              <div v-if="story.summary" style="margin-bottom: 8px;">
+                <span style="font-weight: 600; color: #335599; font-size: 13px;">【故事梗概】</span>
+                <p style="margin: 4px 0 8px; text-indent: 2em; line-height: 1.6; color: var(--text-secondary);">{{ story.summary }}</p>
+              </div>
+
+              <!-- 关键词标签 -->
+              <div v-if="story.tags && story.tags.length" style="margin-bottom: 12px;">
+                <span style="font-weight: 600; color: #335599; font-size: 13px;">【关键词标签】</span>
+                <p style="margin: 4px 0 8px; color: var(--text-muted); font-size: 13px;">{{ story.tags.join('、') }}</p>
+              </div>
+
+              <!-- 故事内容（Markdown 渲染） -->
+              <div v-if="story.content" class="story-content-preview" v-html="renderMarkdown(story.content)"></div>
+            </div>
+            <div v-if="!previewData.stories || !previewData.stories.length" class="text-muted text-sm">
+              暂无故事内容。
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+    <!-- Restart Button -->
+    <div style="text-align: center; margin-top: 20px;">
+      <button class="btn btn-outline" @click="$emit('restart')">
+        <i class="fas fa-rotate-right"></i>
+        再来一次
+      </button>
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { ref, computed, onMounted } from 'vue'
+import { getExportList, previewExport, downloadOne, downloadAll } from '../api'
+
+// Markdown 渲染（简易版，支持 ## 标题、> 引用、- 列表）
+function renderMarkdown(text) {
+  if (!text) return ''
+  return text
+    .split('\n')
+    .map(line => {
+      const trimmed = line.trim()
+      if (!trimmed) return '<br>'
+      if (trimmed.startsWith('## ')) return `<h4 style="font-size:14px;font-weight:700;margin:12px 0 6px;">${trimmed.slice(3)}</h4>`
+      if (trimmed.startsWith('### ')) return `<h5 style="font-size:13px;font-weight:600;margin:8px 0 4px;">${trimmed.slice(4)}</h5>`
+      if (trimmed.startsWith('> ')) return `<blockquote style="border-left:3px solid #ddd;margin:6px 0;padding:4px 12px;color:#666;font-style:italic;">${trimmed.slice(2)}</blockquote>`
+      if (trimmed.startsWith('- ')) return `<div style="padding-left:16px;margin:2px 0;">• ${trimmed.slice(2)}</div>`
+      return `<p style="text-indent:2em;line-height:1.8;margin:4px 0;">${trimmed}</p>`
+    })
+    .join('')
+}
+import { useToast } from '../composables/useToast'
+
+const emit = defineEmits(['next', 'restart'])
+
+const { showToast } = useToast()
+
+const exportList = ref([])
+const loading = ref(false)
+const selectedPersonId = ref(null)
+const previewData = ref(null)
+const previewLoading = ref(false)
+const showMaterialIdx = ref(null)
+
+function toggleMaterial(idx) {
+  showMaterialIdx.value = showMaterialIdx.value === idx ? null : idx
+}
+
+function formatMaterial(raw) {
+  try {
+    const obj = typeof raw === 'string' ? JSON.parse(raw) : raw
+    return JSON.stringify(obj, null, 2)
+  } catch {
+    return raw
+  }
+}
+
+function triggerDownload(blob, filename) {
+  const url = window.URL.createObjectURL(blob)
+  const a = document.createElement('a')
+  a.href = url
+  a.download = filename
+  document.body.appendChild(a)
+  a.click()
+  document.body.removeChild(a)
+  window.URL.revokeObjectURL(url)
+}
+
+async function fetchExportList() {
+  loading.value = true
+  try {
+    const res = await getExportList()
+    exportList.value = res.data || []
+  } catch (e) {
+    showToast('获取导出列表失败: ' + (e.response?.data?.detail || e.message), 'danger')
+  } finally {
+    loading.value = false
+  }
+}
+
+async function handlePreview(personId) {
+  selectedPersonId.value = personId
+  previewLoading.value = true
+  previewData.value = null
+  try {
+    const res = await previewExport(personId)
+    previewData.value = res.data
+  } catch (e) {
+    showToast('加载预览失败: ' + (e.response?.data?.detail || e.message), 'danger')
+  } finally {
+    previewLoading.value = false
+  }
+}
+
+async function handleDownload(personId, personName) {
+  try {
+    const res = await downloadOne(personId)
+    triggerDownload(res.data, `${personName}.docx`)
+    showToast(`${personName} 的文档已开始下载`, 'success')
+  } catch (e) {
+    showToast('下载失败: ' + (e.response?.data?.detail || e.message), 'danger')
+  }
+}
+
+const selectedIndex = computed(() => {
+  if (!selectedPersonId.value) return -1
+  return exportList.value.findIndex(item => item.person_id === selectedPersonId.value)
+})
+
+const hasPrev = computed(() => selectedIndex.value > 0)
+const hasNext = computed(() => selectedIndex.value >= 0 && selectedIndex.value < exportList.value.length - 1)
+
+function navigatePreview(direction) {
+  const newIndex = selectedIndex.value + direction
+  if (newIndex >= 0 && newIndex < exportList.value.length) {
+    const item = exportList.value[newIndex]
+    if (item.story_count) {
+      handlePreview(item.person_id)
+    }
+  }
+}
+
+async function handleDownloadAll() {
+  try {
+    const res = await downloadAll()
+    triggerDownload(res.data, '全部故事文档.zip')
+    showToast('全部文档已开始下载', 'success')
+  } catch (e) {
+    showToast('批量下载失败: ' + (e.response?.data?.detail || e.message), 'danger')
+  }
+}
+
+// Expose handleDownloadAll so the parent can call it via the BottomNav "全部导出" button
+defineExpose({ handleDownloadAll })
+
+onMounted(() => {
+  fetchExportList()
+})
+</script>
+
+<style scoped>
+.export-item.active {
+  border-color: var(--primary);
+  background: var(--primary-bg);
+}
+
+.export-item-name {
+  font-size: 14px;
+  font-weight: 600;
+  color: var(--text);
+}
+
+.export-item-meta {
+  font-size: 12px;
+  color: var(--text-muted);
+  margin-top: 2px;
+}
+
+.file-action-btn:disabled {
+  opacity: 0.4;
+  cursor: not-allowed;
+}
+.file-action-btn:disabled:hover {
+  border-color: transparent;
+  background: transparent;
+  color: var(--text-muted);
+}
+</style>

frontend/src/views/ExtractView.vue

@@ -0,0 +1,410 @@
+<template>
+  <div class="step-page active">
+    <!-- Progress Card -->
+    <div class="card" v-if="!isCompleted && !hasError">
+      <div class="card-header">
+        <div class="card-title">
+          <i class="fas fa-brain" style="width:32px;height:32px;border-radius:8px;display:flex;align-items:center;justify-content:center;font-size:14px;background:#ede9fe;color:#7c3aed;"></i>
+          AI 正在工作中
+        </div>
+        <span class="card-badge" :style="{ fontSize: '18px', fontWeight: '800', color: isCompleted ? 'var(--success)' : 'var(--primary)' }">
+          {{ progress }}%
+        </span>
+      </div>
+
+      <!-- Progress Bar -->
+      <div class="progress-bar" style="height: 10px; border-radius: 5px;">
+        <div class="progress-fill" :style="{ width: progress + '%', transition: 'width 1s ease' }"></div>
+      </div>
+
+      <!-- Current File Info -->
+      <div style="margin-top: 20px;">
+        <div style="display: flex; align-items: center; gap: 12px; margin-bottom: 14px;">
+          <div class="status-dot" :class="isCompleted ? 'done' : 'running'"></div>
+          <div style="flex: 1;">
+            <div style="font-size: 14px; font-weight: 600; color: var(--text);">{{ displayFile }}</div>
+            <div style="font-size: 12px; color: var(--text-muted); margin-top: 2px;">{{ displayAction }}</div>
+          </div>
+        </div>
+
+        <!-- Phase Progress -->
+        <div v-if="phase" style="margin-bottom: 16px;">
+          <div style="display: flex; justify-content: space-between; align-items: center; margin-bottom: 6px;">
+            <span style="font-size: 12px; font-weight: 600; color: var(--text);">{{ phaseLabel }}</span>
+            <span style="font-size: 12px; color: var(--text-muted);">{{ phasePercent }}%</span>
+          </div>
+          <div class="progress-bar" style="height: 4px; border-radius: 2px;">
+            <div class="progress-fill" :style="{ width: phasePercent + '%', transition: 'width 0.5s ease', background: 'var(--primary)' }"></div>
+          </div>
+          <div v-if="phaseDetail" style="font-size: 11px; color: var(--text-muted); margin-top: 4px;">{{ phaseDetail }}</div>
+        </div>
+
+        <!-- Stats Grid -->
+        <div style="display: grid; grid-template-columns: repeat(3, 1fr); gap: 10px; margin-top: 16px;">
+          <div style="background: var(--border-light); border-radius: 10px; padding: 12px 14px; text-align: center;">
+            <div style="font-size: 11px; color: var(--text-muted); margin-bottom: 4px;">文件进度</div>
+            <div style="font-size: 13px; font-weight: 700; color: var(--text);">{{ filesCompleted }} / {{ totalFiles || '...' }}</div>
+          </div>
+          <div style="background: var(--border-light); border-radius: 10px; padding: 12px 14px; text-align: center;">
+            <div style="font-size: 11px; color: var(--text-muted); margin-bottom: 4px;">当前阶段</div>
+            <div style="font-size: 13px; font-weight: 700;" :style="{ color: phase ? 'var(--primary)' : 'var(--text)' }">
+              {{ phaseLabelShort || '等待中' }}
+            </div>
+          </div>
+          <div style="background: var(--border-light); border-radius: 10px; padding: 12px 14px; text-align: center;">
+            <div style="font-size: 11px; color: var(--text-muted); margin-bottom: 4px;">故事发现</div>
+            <div style="font-size: 13px; font-weight: 700;" :style="{ color: storiesFound > 0 ? 'var(--success)' : 'var(--text)' }">
+              {{ storiesFound }} 个
+            </div>
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <!-- Error Card -->
+    <div class="card" v-if="hasError">
+      <div class="card-header">
+        <div class="card-title">
+          <i class="fas fa-circle-exclamation" style="width:32px;height:32px;border-radius:8px;display:flex;align-items:center;justify-content:center;font-size:14px;background:#fee2e2;color:#ef4444;"></i>
+          无法提取故事
+        </div>
+      </div>
+      <p class="text-sm" style="color: var(--danger); margin-bottom: 16px;">{{ errorMessage }}</p>
+      <p class="text-sm" style="color: var(--text-muted); margin-bottom: 16px;">
+        请检查上传的文件是否包含学员与老师的对话记录，而非纯课程内容。
+      </p>
+      <div style="display: flex; gap: 12px;">
+        <button class="btn btn-outline" @click="$emit('go-to', 'upload')">
+          <i class="fas fa-arrow-left"></i>
+          返回重选文件
+        </button>
+        <button class="btn btn-primary" @click="retryExtraction">
+          <i class="fas fa-rotate-right"></i>
+          重新提取
+        </button>
+      </div>
+    </div>
+
+    <!-- Results Card -->
+    <div class="card" v-if="isCompleted">
+      <div class="card-header">
+        <div class="card-title">
+          <i class="icon-magic"><i class="fas fa-list"></i></i>
+          已提取的故事
+        </div>
+        <span class="card-badge">共 {{ stories.length }} 个故事</span>
+      </div>
+
+      <div v-if="stories.length === 0" style="text-align: center; padding: 32px 0;">
+        <i class="fas fa-inbox" style="font-size: 28px; display: block; margin-bottom: 12px; color: var(--border);"></i>
+        <p class="text-muted text-sm" style="margin-bottom: 16px;">未发现任何故事。</p>
+        <p class="text-muted text-sm" style="margin-bottom: 16px;">
+          可能原因：文件中没有学员的深度对话，或 AI 服务在提取过程中出错。
+        </p>
+        <div style="display: flex; gap: 12px; justify-content: center;">
+          <button class="btn btn-outline" @click="$emit('go-to', 'upload')">
+            <i class="fas fa-arrow-left"></i>
+            返回重选文件
+          </button>
+          <button class="btn btn-primary" @click="retryExtraction">
+            <i class="fas fa-rotate-right"></i>
+            重新提取
+          </button>
+        </div>
+      </div>
+
+      <div v-for="story in stories" :key="story.id" class="story-card">
+        <div style="display: flex; justify-content: space-between; align-items: start;">
+          <div style="flex: 1; min-width: 0;">
+            <div class="story-title">{{ story.title }}</div>
+            <div class="story-tag" :class="story.match_status === 'matched' ? 'matched' : 'pending'">
+              <i class="fas" :class="story.match_status === 'matched' ? 'fa-check' : 'fa-clock'"></i>
+              {{ story.match_status === 'matched' ? '已匹配' : '待匹配' }}
+            </div>
+          </div>
+          <div class="story-actions">
+            <button class="btn btn-outline btn-sm" @click="viewStory(story)" title="查看">
+              <i class="fas fa-eye"></i>
+            </button>
+            <button class="btn btn-ghost btn-sm" style="color: var(--danger);" @click="handleDeleteStory(story)" title="删除">
+              <i class="fas fa-trash-alt"></i>
+            </button>
+          </div>
+        </div>
+        <div class="story-summary">{{ story.summary }}</div>
+        <div class="story-source">
+          <i class="fas fa-file-alt"></i>
+          {{ formatSource(story) }}
+        </div>
+      </div>
+    </div>
+
+    <!-- Story Detail Modal -->
+    <div v-if="viewingStory" class="modal-overlay" @click.self="viewingStory = null">
+      <div class="card" style="max-width: 640px; width: 90%; max-height: 80vh; overflow-y: auto; position: relative;">
+        <div class="card-header">
+          <div class="card-title">{{ viewingStory.title }}</div>
+          <button class="btn btn-ghost btn-sm" @click="viewingStory = null" style="font-size: 18px; color: var(--text-muted);">&times;</button>
+        </div>
+        <div style="margin-bottom: 12px;">
+          <span v-for="tag in (viewingStory.tags_json || [])" :key="tag" style="display: inline-block; background: var(--border-light); border-radius: 12px; padding: 2px 10px; font-size: 12px; color: var(--text-muted); margin-right: 6px; margin-bottom: 4px;">{{ tag }}</span>
+        </div>
+        <div style="font-size: 13px; color: var(--text-muted); margin-bottom: 16px;">{{ viewingStory.summary }}</div>
+        <div v-html="formattedContent" style="font-size: 14px; line-height: 1.7; color: var(--text);"></div>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { ref, computed, inject, onMounted, onUnmounted } from 'vue'
+import { useRouter } from 'vue-router'
+import { startExtraction, streamExtractionStatus, getStories, deleteStory, getExtractionStatus } from '../api'
+import { useToast } from '../composables/useToast'
+
+const router = useRouter()
+const { showToast } = useToast()
+
+const setExtracting = inject('setExtracting', () => {})
+
+defineEmits(['next', 'prev', 'go-to'])
+
+const taskId = ref(null)
+
+// Progress state
+const progress = ref(0)
+const currentFile = ref('')
+const currentAction = ref('')
+const filesCompleted = ref(0)
+const totalFiles = ref(0)
+const storiesFound = ref(0)
+const phase = ref('')
+const phasePercent = ref(0)
+const phaseDetail = ref('')
+
+// Results state
+const isCompleted = ref(false)
+const hasError = ref(false)
+const errorMessage = ref('')
+const stories = ref([])
+const viewingStory = ref(null)
+
+const formattedContent = computed(() => {
+  if (!viewingStory.value) return ''
+  return viewingStory.value.content || ''
+})
+
+let eventSource = null
+
+const phaseLabels = {
+  preprocess: '预处理分析',
+  identify: '识别故事',
+  chunk: '切片提取',
+  merge: '素材重组',
+  generate: '生成故事',
+}
+
+const phaseLabel = computed(() => phaseLabels[phase.value] || '')
+const phaseLabelShort = computed(() => phaseLabels[phase.value] || '')
+
+const displayFile = computed(() => {
+  if (isCompleted.value) return '全部处理完成！'
+  if (currentFile.value) return '正在处理：' + currentFile.value
+  return '准备中...'
+})
+
+const displayAction = computed(() => {
+  if (isCompleted.value) return `共处理 ${totalFiles.value} 份文字稿，发现 ${storiesFound.value} 个故事`
+  if (currentAction.value) return currentAction.value
+  return '初始化...'
+})
+
+function connectSSE() {
+  if (eventSource) {
+    eventSource.close()
+  }
+
+  eventSource = streamExtractionStatus(taskId.value)
+
+  eventSource.addEventListener('progress', (event) => {
+    try {
+      const data = JSON.parse(event.data)
+
+      currentFile.value = data.current_file || ''
+      currentAction.value = data.action || ''
+      filesCompleted.value = data.files_completed || 0
+      totalFiles.value = data.files_total || 0
+      storiesFound.value = data.stories_found || 0
+      progress.value = Math.round(data.percent || 0)
+      phase.value = data.phase || ''
+      phasePercent.value = Math.round(data.phase_percent || 0)
+      phaseDetail.value = data.phase_detail || ''
+
+      if (data.status === 'completed') {
+        isCompleted.value = true
+        progress.value = 100
+        phasePercent.value = 100
+        setExtracting(false)
+        closeSSE()
+        loadStories()
+      } else if (data.status === 'error' || data.error) {
+        hasError.value = true
+        errorMessage.value = data.error || data.action || '提取过程中发生未知错误'
+        setExtracting(false)
+        closeSSE()
+      } else if (data.status === 'stopped') {
+        hasError.value = true
+        errorMessage.value = '提取已停止'
+        setExtracting(false)
+        closeSSE()
+      }
+    } catch (e) {
+      console.error('Failed to parse SSE data:', e)
+    }
+  })
+
+  eventSource.onerror = () => {
+    console.warn('SSE connection error')
+  }
+}
+
+function closeSSE() {
+  if (eventSource) {
+    eventSource.close()
+    eventSource = null
+  }
+}
+
+async function loadStories() {
+  try {
+    const res = await getStories()
+    stories.value = res.data || []
+  } catch (e) {
+    console.error('Failed to load stories:', e)
+    showToast('加载故事列表失败', 'warning')
+  }
+}
+
+async function handleDeleteStory(story) {
+  try {
+    await deleteStory(story.id)
+    stories.value = stories.value.filter(s => s.id !== story.id)
+    storiesFound.value = Math.max(0, storiesFound.value - 1)
+    showToast('故事已删除', 'warning')
+  } catch (e) {
+    console.error('Failed to delete story:', e)
+    showToast('删除故事失败', 'warning')
+  }
+}
+
+function viewStory(story) {
+  // 解析 tags JSON
+  try {
+    story.tags_json = typeof story.tags === 'string' ? JSON.parse(story.tags) : (story.tags || [])
+  } catch {
+    story.tags_json = []
+  }
+  viewingStory.value = story
+}
+
+async function retryExtraction() {
+  hasError.value = false
+  errorMessage.value = ''
+  isCompleted.value = false
+  progress.value = 0
+  currentFile.value = ''
+  currentAction.value = ''
+  filesCompleted.value = 0
+  totalFiles.value = 0
+  storiesFound.value = 0
+  phase.value = ''
+  phasePercent.value = 0
+  phaseDetail.value = ''
+  stories.value = []
+
+  await initiateExtraction()
+}
+
+async function initiateExtraction() {
+  setExtracting(true)
+  try {
+    const res = await startExtraction(taskId.value)
+    if (res.data?.task_id) {
+      taskId.value = res.data.task_id
+      localStorage.setItem('extraction_task_id', res.data.task_id)
+    }
+    connectSSE()
+  } catch (e) {
+    if (e.response && e.response.status === 400) {
+      showToast('提取任务正在进行中', 'warning')
+      connectSSE()
+    } else {
+      hasError.value = true
+      errorMessage.value = e.response?.data?.detail || '无法启动提取任务'
+      setExtracting(false)
+      showToast('启动提取失败', 'warning')
+    }
+  }
+}
+
+function formatSource(story) {
+  const parts = []
+  if (story.source_lines) {
+    try {
+      const lines = typeof story.source_lines === 'string' ? JSON.parse(story.source_lines) : story.source_lines
+      if (Array.isArray(lines) && lines.length > 0) {
+        const min = Math.min(...lines)
+        const max = Math.max(...lines)
+        parts.push(`第 ${min.toLocaleString()}-${max.toLocaleString()} 行`)
+      }
+    } catch (e) {
+      // ignore
+    }
+  }
+  if (story.duration_minutes) {
+    parts.push(`约 ${Math.round(story.duration_minutes)} 分钟`)
+  }
+  return parts.join(' · ') || '未知来源'
+}
+
+onMounted(async () => {
+  const savedTaskId = localStorage.getItem('extraction_task_id')
+  if (savedTaskId) {
+    taskId.value = savedTaskId
+    try {
+      const { data } = await getExtractionStatus(savedTaskId)
+      if (data.status === 'completed') {
+        isCompleted.value = true
+        progress.value = 100
+        await loadStories()
+        return
+      }
+      if (data.status === 'running') {
+        setExtracting(true)
+        connectSSE()
+        return
+      }
+    } catch (e) {
+      // 任务不存在，清除并重新开始
+    }
+  }
+  await initiateExtraction()
+})
+
+onUnmounted(() => {
+  closeSSE()
+  setExtracting(false)
+})
+</script>
+
+<style scoped>
+.modal-overlay {
+  position: fixed;
+  top: 0; left: 0; right: 0; bottom: 0;
+  background: rgba(0,0,0,0.4);
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  z-index: 1000;
+}
+</style>

frontend/src/views/MatchView.vue

@@ -0,0 +1,288 @@
+<template>
+  <div class="step-page active">
+    <!-- Header Card -->
+    <div class="card">
+      <div class="card-header">
+        <div class="card-title">
+          <i class="fas fa-link icon-link"></i>
+          故事匹配
+        </div>
+        <div class="card-badge" v-if="stories.length">
+          {{ matchedStories.length }} / {{ stories.length }} 已匹配
+        </div>
+      </div>
+      <p class="text-muted text-sm">将提取的故事与讲述者进行一一对应。选择左侧故事，选择右侧讲述者，点击中间箭头完成匹配。</p>
+    </div>
+
+    <!-- Match Panel: 3 columns -->
+    <div class="match-panel">
+      <!-- Left Column: Pending Stories -->
+      <div class="card" style="margin-bottom: 0;">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="fas fa-book icon-magic"></i>
+            待匹配故事
+          </div>
+          <div class="card-badge" v-if="pendingStories.length">
+            {{ pendingStories.length }} 条
+          </div>
+        </div>
+        <div class="scrollable">
+          <template v-if="pendingStories.length">
+            <div
+              v-for="story in pendingStories"
+              :key="story.id"
+              class="story-card"
+              :class="{ 'match-highlight': selectedStory?.id === story.id }"
+              @click="selectStory(story)"
+            >
+              <div class="story-title">{{ story.title }}</div>
+              <div class="story-tag pending">
+                <i class="fas fa-clock"></i> 待匹配
+              </div>
+              <div class="story-summary" v-if="story.summary">
+                {{ truncateText(story.summary, 80) }}
+              </div>
+              <div class="story-source">
+                <i class="fas fa-file-lines"></i>
+                {{ story.speaker_nickname || '未知讲述者' }}
+                <span v-if="story.duration_minutes"> · {{ story.duration_minutes }}分钟</span>
+              </div>
+            </div>
+          </template>
+          <p v-else class="text-muted text-sm" style="text-align: center; padding: 24px 0;">
+            <i class="fas fa-check-circle" style="color: var(--success); margin-right: 4px;"></i>
+            所有故事均已匹配完成
+          </p>
+        </div>
+      </div>
+
+      <!-- Center Column: Arrow Button -->
+      <div class="match-center-col">
+        <button
+          class="match-arrow-btn"
+          title="匹配"
+          :disabled="!selectedStory || !selectedPerson || matching"
+          @click="handleMatch"
+        >
+          <i class="fas fa-arrow-right" v-if="!matching"></i>
+          <i class="fas fa-spinner fa-spin" v-else></i>
+        </button>
+        <span class="match-hint">
+          <template v-if="!selectedStory && !selectedPerson">选择故事后<br>点击匹配</template>
+          <template v-else-if="!selectedPerson">请选择<br>讲述者</template>
+          <template v-else-if="!selectedStory">请选择<br>故事</template>
+          <template v-else-if="matching">匹配中...</template>
+          <template v-else>点击匹配</template>
+        </span>
+      </div>
+
+      <!-- Right Column: Persons -->
+      <div class="card" style="margin-bottom: 0;">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="fas fa-user-group icon-person"></i>
+            讲述者
+          </div>
+          <div class="card-badge">{{ persons.length + 1 }} 人</div>
+        </div>
+        <div class="scrollable">
+          <!-- 默认「暂无」选项，始终置顶 -->
+          <div
+            class="person-card"
+            :class="{ selected: selectedPerson?.id === 'unknown' }"
+            @click="selectPerson({ id: 'unknown', name: '暂无' })"
+          >
+            <div class="person-avatar" style="background: linear-gradient(135deg, #f1f5f9, #e2e8f0); color: var(--text-muted);">
+              <i class="fas fa-question" style="font-size: 14px;"></i>
+            </div>
+            <div>
+              <div class="person-name">暂无</div>
+              <div class="person-meta">{{ getUnknownMatchCount() }}</div>
+            </div>
+          </div>
+          <!-- 真实讲述者列表 -->
+          <template v-if="persons.length">
+            <div
+              v-for="person in persons"
+              :key="person.id"
+              class="person-card"
+              :class="{ selected: selectedPerson?.id === person.id }"
+              @click="selectPerson(person)"
+            >
+              <div class="person-avatar">{{ person.name?.charAt(0) || '?' }}</div>
+              <div>
+                <div class="person-name">{{ person.name }}</div>
+                <div class="person-meta">{{ getPersonMatchCount(person.id) }}</div>
+              </div>
+            </div>
+          </template>
+          <p v-else class="text-muted text-sm" style="text-align: center; padding: 16px 0;">
+            暂无讲述者信息
+          </p>
+        </div>
+      </div>
+    </div>
+
+    <!-- Matched Stories Section -->
+    <div class="card mt-4" v-if="matchedStories.length">
+      <div class="card-header">
+        <div class="card-title">
+          <i class="fas fa-check-double icon-export"></i>
+          已匹配故事
+        </div>
+        <div class="card-badge">{{ matchedStories.length }} 条</div>
+      </div>
+      <div class="scrollable" style="max-height: 40vh;">
+        <div
+          v-for="story in matchedStories"
+          :key="story.id"
+          class="story-card"
+        >
+          <div class="story-title">{{ story.title }}</div>
+          <div class="story-tag matched">
+            <i class="fas fa-check"></i> 已匹配
+          </div>
+          <div class="story-summary" v-if="story.summary">
+            {{ truncateText(story.summary, 80) }}
+          </div>
+          <div class="story-source">
+            <i class="fas fa-user"></i>
+            {{ getPersonName(story.person_id) }}
+          </div>
+          <div class="story-actions">
+            <button
+              class="btn btn-sm btn-outline"
+              style="color: var(--danger); border-color: #fecaca;"
+              @click="handleUnmatch(story)"
+              :disabled="unmatchingId === story.id"
+            >
+              <i class="fas fa-link-slash" v-if="unmatchingId !== story.id"></i>
+              <i class="fas fa-spinner fa-spin" v-else></i>
+              取消匹配
+            </button>
+          </div>
+        </div>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { ref, computed, onMounted } from 'vue'
+import { getStories, getPersons, createMatch, removeMatch } from '../api'
+import { useToast } from '../composables/useToast'
+
+const emit = defineEmits(['next', 'prev', 'go-to'])
+
+const { showToast } = useToast()
+
+// Data
+const stories = ref([])
+const persons = ref([])
+const selectedStory = ref(null)
+const selectedPerson = ref(null)
+const matching = ref(false)
+const unmatchingId = ref(null)
+const loading = ref(false)
+
+// Computed
+const pendingStories = computed(() =>
+  stories.value.filter(s => s.match_status === 'pending')
+)
+
+const matchedStories = computed(() =>
+  stories.value.filter(s => s.match_status === 'matched')
+)
+
+// Methods
+function truncateText(text, maxLen) {
+  if (!text) return ''
+  return text.length > maxLen ? text.slice(0, maxLen) + '...' : text
+}
+
+function getPersonName(personId) {
+  if (personId === 'unknown') return '暂无'
+  const person = persons.value.find(p => p.id === personId)
+  return person?.name || '未知讲述者'
+}
+
+function getPersonMatchCount(personId) {
+  const count = stories.value.filter(s => s.person_id === personId && s.match_status === 'matched').length
+  return count > 0 ? `已匹配 ${count} 个故事` : '未匹配'
+}
+
+function getUnknownMatchCount() {
+  const count = stories.value.filter(s => s.person_id === 'unknown' && s.match_status === 'matched').length
+  return count > 0 ? `已匹配 ${count} 个故事` : '未知讲述者可用'
+}
+
+function selectStory(story) {
+  selectedStory.value = selectedStory.value?.id === story.id ? null : story
+}
+
+function selectPerson(person) {
+  selectedPerson.value = selectedPerson.value?.id === person.id ? null : person
+}
+
+async function fetchAllData() {
+  loading.value = true
+  try {
+    const [storiesRes, personsRes] = await Promise.all([
+      getStories('all'),
+      getPersons()
+    ])
+    stories.value = storiesRes.data || []
+    persons.value = personsRes.data || []
+  } catch (err) {
+    console.error('Failed to fetch data:', err)
+    showToast('加载数据失败，请稍后重试', 'error')
+  } finally {
+    loading.value = false
+  }
+}
+
+async function handleMatch() {
+  if (!selectedStory.value || !selectedPerson.value) {
+    showToast('请同时选择一个故事和一个讲述者', 'warning')
+    return
+  }
+  if (matching.value) return
+
+  matching.value = true
+  try {
+    await createMatch(selectedStory.value.id, selectedPerson.value.id)
+    showToast(`已将「${selectedStory.value.title}」匹配给「${selectedPerson.value.name}」`, 'success')
+    selectedStory.value = null
+    selectedPerson.value = null
+    await fetchAllData()
+  } catch (err) {
+    console.error('Match failed:', err)
+    const msg = err.response?.data?.detail || '匹配失败，请稍后重试'
+    showToast(msg, 'error')
+  } finally {
+    matching.value = false
+  }
+}
+
+async function handleUnmatch(story) {
+  if (unmatchingId.value) return
+
+  unmatchingId.value = story.id
+  try {
+    await removeMatch(story.id)
+    showToast(`已取消「${story.title}」的匹配`, 'success')
+    await fetchAllData()
+  } catch (err) {
+    console.error('Unmatch failed:', err)
+    const msg = err.response?.data?.detail || '取消匹配失败，请稍后重试'
+    showToast(msg, 'error')
+  } finally {
+    unmatchingId.value = null
+  }
+}
+
+onMounted(() => {
+  fetchAllData()
+})
+</script>

frontend/src/views/UploadView.vue

@@ -0,0 +1,292 @@
+<template>
+  <div class="step-page active">
+    <div class="two-col">
+      <!-- Left Column: Transcripts -->
+      <div class="card">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="icon-upload"><i class="fas fa-file-alt"></i></i>
+            直播文字稿
+          </div>
+          <span class="card-badge">DOCX 格式</span>
+        </div>
+        <div
+          class="upload-zone"
+          :class="{ 'is-dragging': transcriptDragging }"
+          @click="triggerTranscriptInput"
+          @dragover.prevent="transcriptDragging = true"
+          @dragleave.prevent="transcriptDragging = false"
+          @drop.prevent="handleTranscriptDrop"
+        >
+          <i class="fas fa-cloud-upload-alt upload-icon"></i>
+          <p>拖拽文件到此处，或点击上传</p>
+          <p class="hint">支持批量上传 · 仅限 .docx 格式</p>
+        </div>
+        <input
+          ref="transcriptInputRef"
+          type="file"
+          accept=".docx"
+          multiple
+          style="display: none"
+          @change="handleTranscriptSelect"
+        />
+        <div class="file-list">
+          <div v-if="transcriptUploading" class="upload-progress">
+            <div class="progress-bar">
+              <div class="progress-fill" :style="{ width: transcriptProgress + '%' }"></div>
+            </div>
+            <p class="text-sm text-muted mt-2">正在上传...</p>
+          </div>
+          <div
+            v-for="item in transcripts"
+            :key="item.id"
+            class="file-item"
+          >
+            <div class="file-icon docx">
+              <i class="fas fa-file-word"></i>
+            </div>
+            <div class="file-info">
+              <div class="file-name">{{ item.filename }}</div>
+              <div class="file-meta">
+                {{ formatLineCount(item.line_count) }} 行 · {{ formatFileSize(item.file_size) }}
+              </div>
+            </div>
+            <button
+              class="file-action-btn danger"
+              title="删除"
+              @click="handleDeleteTranscript(item.id)"
+            >
+              <i class="fas fa-trash-alt"></i>
+            </button>
+          </div>
+        </div>
+      </div>
+
+      <!-- Right Column: Persons -->
+      <div class="card">
+        <div class="card-header">
+          <div class="card-title">
+            <i class="icon-person"><i class="fas fa-user"></i></i>
+            讲述者信息
+          </div>
+          <span class="card-badge">一人一个 DOCX</span>
+        </div>
+        <div
+          class="upload-zone"
+          :class="{ 'is-dragging': personDragging }"
+          @click="triggerPersonInput"
+          @dragover.prevent="personDragging = true"
+          @dragleave.prevent="personDragging = false"
+          @drop.prevent="handlePersonDrop"
+        >
+          <i class="fas fa-user-plus upload-icon"></i>
+          <p>拖拽文件到此处，或点击上传</p>
+          <p class="hint">每个讲述者一个 DOCX · 包含个人信息和照片</p>
+        </div>
+        <input
+          ref="personInputRef"
+          type="file"
+          accept=".docx"
+          multiple
+          style="display: none"
+          @change="handlePersonSelect"
+        />
+        <div class="file-list">
+          <div v-if="personUploading" class="upload-progress">
+            <div class="progress-bar">
+              <div class="progress-fill" :style="{ width: personProgress + '%' }"></div>
+            </div>
+            <p class="text-sm text-muted mt-2">正在上传...</p>
+          </div>
+          <div
+            v-for="item in persons"
+            :key="item.id"
+            class="file-item"
+          >
+            <div class="file-icon person">
+              <i class="fas fa-user"></i>
+            </div>
+            <div class="file-info">
+              <div class="file-name">{{ item.filename }}</div>
+              <div class="file-meta">
+                {{ item.has_photo ? '含照片' : '无照片' }} · {{ formatFileSize(item.file_size) }}
+              </div>
+            </div>
+            <button
+              class="file-action-btn danger"
+              title="删除"
+              @click="handleDeletePerson(item.id)"
+            >
+              <i class="fas fa-trash-alt"></i>
+            </button>
+          </div>
+        </div>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup>
+import { ref, onMounted } from 'vue'
+import {
+  uploadTranscripts,
+  uploadPersons,
+  getTranscripts,
+  getPersons,
+  deleteTranscript,
+  deletePerson,
+} from '../api'
+import { useToast } from '../composables/useToast'
+
+defineEmits(['next', 'prev', 'go-to'])
+
+const { showToast } = useToast()
+
+// ===== Transcript state =====
+const transcripts = ref([])
+const transcriptDragging = ref(false)
+const transcriptUploading = ref(false)
+const transcriptProgress = ref(0)
+const transcriptInputRef = ref(null)
+
+// ===== Person state =====
+const persons = ref([])
+const personDragging = ref(false)
+const personUploading = ref(false)
+const personProgress = ref(0)
+const personInputRef = ref(null)
+
+// ===== Format helpers =====
+function formatFileSize(bytes) {
+  if (!bytes || bytes === 0) return '0 B'
+  if (bytes < 1024) return bytes + ' B'
+  if (bytes < 1024 * 1024) return (bytes / 1024).toFixed(1) + ' KB'
+  return (bytes / (1024 * 1024)).toFixed(1) + ' MB'
+}
+
+function formatLineCount(count) {
+  if (count == null) return '-'
+  return count.toLocaleString('en-US')
+}
+
+// ===== Transcript upload =====
+function triggerTranscriptInput() {
+  transcriptInputRef.value?.click()
+}
+
+function handleTranscriptSelect(e) {
+  const files = Array.from(e.target.files || [])
+  if (files.length > 0) uploadTranscriptFiles(files)
+  e.target.value = ''
+}
+
+function handleTranscriptDrop(e) {
+  transcriptDragging.value = false
+  const files = Array.from(e.dataTransfer.files || []).filter(f =>
+    f.name.toLowerCase().endsWith('.docx')
+  )
+  if (files.length > 0) {
+    uploadTranscriptFiles(files)
+  } else {
+    showToast('请上传 .docx 格式的文件', 'error')
+  }
+}
+
+async function uploadTranscriptFiles(files) {
+  transcriptUploading.value = true
+  transcriptProgress.value = 0
+  try {
+    await uploadTranscripts(files)
+    transcriptProgress.value = 100
+    const res = await getTranscripts()
+    transcripts.value = res.data || []
+  } catch (err) {
+    showToast(err.response?.data?.detail || '上传文字稿失败', 'error')
+  } finally {
+    transcriptUploading.value = false
+    setTimeout(() => { transcriptProgress.value = 0 }, 500)
+  }
+}
+
+async function handleDeleteTranscript(id) {
+  try {
+    await deleteTranscript(id)
+    transcripts.value = transcripts.value.filter(t => t.id !== id)
+  } catch (err) {
+    showToast(err.response?.data?.detail || '删除失败', 'error')
+  }
+}
+
+// ===== Person upload =====
+function triggerPersonInput() {
+  personInputRef.value?.click()
+}
+
+function handlePersonSelect(e) {
+  const files = Array.from(e.target.files || [])
+  if (files.length > 0) uploadPersonFiles(files)
+  e.target.value = ''
+}
+
+function handlePersonDrop(e) {
+  personDragging.value = false
+  const files = Array.from(e.dataTransfer.files || []).filter(f =>
+    f.name.toLowerCase().endsWith('.docx')
+  )
+  if (files.length > 0) {
+    uploadPersonFiles(files)
+  } else {
+    showToast('请上传 .docx 格式的文件', 'error')
+  }
+}
+
+async function uploadPersonFiles(files) {
+  personUploading.value = true
+  personProgress.value = 0
+  try {
+    await uploadPersons(files)
+    personProgress.value = 100
+    const res = await getPersons()
+    persons.value = res.data || []
+  } catch (err) {
+    showToast(err.response?.data?.detail || '上传讲述者信息失败', 'error')
+  } finally {
+    personUploading.value = false
+    setTimeout(() => { personProgress.value = 0 }, 500)
+  }
+}
+
+async function handleDeletePerson(id) {
+  try {
+    await deletePerson(id)
+    persons.value = persons.value.filter(p => p.id !== id)
+  } catch (err) {
+    showToast(err.response?.data?.detail || '删除失败', 'error')
+  }
+}
+
+// ===== Load existing data =====
+onMounted(async () => {
+  try {
+    const [transcriptRes, personRes] = await Promise.all([
+      getTranscripts(),
+      getPersons(),
+    ])
+    transcripts.value = transcriptRes.data || []
+    persons.value = personRes.data || []
+  } catch (err) {
+    showToast('加载文件列表失败', 'error')
+  }
+})
+</script>
+
+<style scoped>
+.upload-zone.is-dragging {
+  border-color: var(--primary);
+  background: var(--primary-bg);
+}
+
+.upload-progress {
+  padding: 12px 0;
+}
+</style>

frontend/src/views/WelcomeView.vue

@@ -0,0 +1,40 @@
+<template>
+  <div class="welcome-page">
+    <div class="welcome-icon">
+      <i class="fas fa-book-open"></i>
+    </div>
+    <h1 class="welcome-title">Story Extractor</h1>
+    <p class="welcome-desc">
+      从口述历史文字稿中，利用 AI 自动提取个人故事，并与讲述者匹配，最终生成结构化文档。
+    </p>
+    <div class="welcome-steps">
+      <div class="welcome-step-item">
+        <div class="welcome-step-icon s1"><i class="fas fa-cloud-upload-alt"></i></div>
+        <span class="welcome-step-label">上传文件</span>
+        <span class="welcome-step-num">Step 1</span>
+      </div>
+      <div class="welcome-step-item">
+        <div class="welcome-step-icon s2"><i class="fas fa-magic"></i></div>
+        <span class="welcome-step-label">提取故事</span>
+        <span class="welcome-step-num">Step 2</span>
+      </div>
+      <div class="welcome-step-item">
+        <div class="welcome-step-icon s3"><i class="fas fa-link"></i></div>
+        <span class="welcome-step-label">匹配确认</span>
+        <span class="welcome-step-num">Step 3</span>
+      </div>
+      <div class="welcome-step-item">
+        <div class="welcome-step-icon s4"><i class="fas fa-file-export"></i></div>
+        <span class="welcome-step-label">预览导出</span>
+        <span class="welcome-step-num">Step 4</span>
+      </div>
+    </div>
+    <button class="btn btn-primary btn-lg" @click="$emit('next')">
+      <i class="fas fa-rocket"></i> 开始使用
+    </button>
+  </div>
+</template>
+
+<script setup>
+defineEmits(['next', 'prev', 'go-to'])
+</script>

frontend/vite.config.js

@@ -0,0 +1,15 @@
+import { defineConfig } from 'vite'
+import vue from '@vitejs/plugin-vue'
+
+export default defineConfig({
+  plugins: [vue()],
+  server: {
+    port: 3000,
+    proxy: {
+      '/api': {
+        target: 'http://localhost:8000',
+        changeOrigin: true,
+      }
+    }
+  }
+})

run.py

@@ -0,0 +1,126 @@
+"""
+故事提取工具 - 入口脚本
+支持开发模式和 PyInstaller 打包模式
+"""
+import os
+import sys
+import shutil
+import socket
+import webbrowser
+import threading
+import time
+
+
+def get_base_dir() -> str:
+    """获取基础目录"""
+    if getattr(sys, 'frozen', False):
+        return os.path.dirname(sys.executable)
+    else:
+        return os.path.dirname(os.path.abspath(__file__))
+
+
+def cleanup_data(base_dir: str):
+    """启动时清理数据库缓存和临时文件"""
+    print("正在清理上次运行的数据...")
+
+    # 实际数据目录（开发模式下在 backend/ 下，打包后在 exe 同级目录下）
+    if getattr(sys, 'frozen', False):
+        data_dir = base_dir
+    else:
+        data_dir = os.path.join(base_dir, "backend")
+
+    # 清理数据库
+    db_dir = os.path.join(data_dir, "data")
+    if os.path.exists(db_dir):
+        for f in os.listdir(db_dir):
+            if f.endswith((".db", ".db-wal", ".db-shm", ".db-journal")):
+                try:
+                    os.remove(os.path.join(db_dir, f))
+                except Exception:
+                    pass
+
+    # 清理上传的文件
+    uploads_dir = os.path.join(data_dir, "uploads")
+    if os.path.exists(uploads_dir):
+        for subdir in ["transcripts", "persons", "photos"]:
+            sub = os.path.join(uploads_dir, subdir)
+            if os.path.exists(sub):
+                shutil.rmtree(sub, ignore_errors=True)
+
+    # 清理导出文件
+    exports_dir = os.path.join(data_dir, "exports")
+    if os.path.exists(exports_dir):
+        shutil.rmtree(exports_dir, ignore_errors=True)
+
+    # 重新创建目录结构
+    for d in [db_dir, uploads_dir, exports_dir]:
+        os.makedirs(d, exist_ok=True)
+    for sub in ["transcripts", "persons", "photos"]:
+        os.makedirs(os.path.join(uploads_dir, sub), exist_ok=True)
+
+    print("清理完成！")
+
+
+def find_free_port(start_port: int = 8000) -> int:
+    """找到一个可用端口"""
+    for port in range(start_port, start_port + 100):
+        try:
+            with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+                s.bind(("127.0.0.1", port))
+                return port
+        except OSError:
+            continue
+    return start_port
+
+
+def open_browser(port: int):
+    """延迟打开浏览器"""
+    time.sleep(2)
+    url = f"http://localhost:{port}"
+    print(f"\n正在打开浏览器: {url}")
+    webbrowser.open(url)
+
+
+def main():
+    base_dir = get_base_dir()
+
+    # 1. 清理数据
+    cleanup_data(base_dir)
+
+    # 2. 找可用端口
+    port = find_free_port(8000)
+    if port != 8000:
+        print(f"端口 8000 被占用，使用端口 {port}")
+
+    # 3. 启动服务
+    print(f"\n{'='*50}")
+    print(f"  故事提取工具")
+    print(f"  访问地址: http://localhost:{port}")
+    print(f"  按 Ctrl+C 停止服务")
+    print(f"{'='*50}\n")
+
+    # 延迟打开浏览器（不阻塞主线程）
+    threading.Thread(target=open_browser, args=(port,), daemon=True).start()
+
+    # 启动 uvicorn
+    os.environ["APP_PORT"] = str(port)
+
+    if getattr(sys, 'frozen', False):
+        # 打包模式：直接 import 并运行
+        import uvicorn
+        # 把 backend 目录加到 path，让 app 包能被找到
+        backend_dir = os.path.join(base_dir, "backend")
+        if backend_dir not in sys.path:
+            sys.path.insert(0, backend_dir)
+        uvicorn.run("app.main:app", host="0.0.0.0", port=port, log_level="info")
+    else:
+        # 开发模式
+        import uvicorn
+        backend_dir = os.path.join(base_dir, "backend")
+        if backend_dir not in sys.path:
+            sys.path.insert(0, backend_dir)
+        uvicorn.run("app.main:app", host="0.0.0.0", port=port, log_level="info")
+
+
+if __name__ == "__main__":
+    main()

test_pipeline.py

@@ -0,0 +1,733 @@
+"""
+故事提取流程 v4 - 逐步测试脚本
+用法：python test_pipeline.py <docx文件路径> [步骤编号]
+  步骤编号：1=预处理, 2=判断故事, 3=切片提取, 4=重组, 5=生成故事, all=全部执行
+"""
+import sys
+import os
+import json
+import asyncio
+import time
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "backend"))
+
+from app.services.docx_parser import parse_transcript
+from app.services.llm_client import llm_client, _parse_json
+
+TEST_DIR = os.path.join(os.path.dirname(__file__), "backend", "uploads", "test")
+
+
+def _save_json(filepath, data):
+    """保存 JSON 到文件"""
+    os.makedirs(os.path.dirname(filepath), exist_ok=True)
+    with open(filepath, "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+
+def _save_text(filepath, text):
+    """保存文本到文件"""
+    os.makedirs(os.path.dirname(filepath), exist_ok=True)
+    with open(filepath, "w", encoding="utf-8") as f:
+        f.write(text)
+
+
+# ============================================================
+# 第一步：代码预处理（v3：时间范围截取，允许重叠）
+# ============================================================
+def step1_preprocess(lines, output_dir, save=True):
+    """代码预处理：识别老师 → 过滤杂音 → 按时间范围截取每个学员的全部内容"""
+    print("\n" + "=" * 60)
+    print("第一步：代码预处理（v3：时间范围截取）")
+    print("=" * 60)
+
+    # 1.1 统计每个说话人
+    print(f"总行数: {len(lines)}")
+    speaker_stats = {}
+    for i, line in enumerate(lines):
+        speaker = line.get("speaker", "")
+        if not speaker:
+            continue
+        if speaker not in speaker_stats:
+            speaker_stats[speaker] = {"lines": 0, "chars": 0, "first": i, "last": i}
+        speaker_stats[speaker]["lines"] += 1
+        speaker_stats[speaker]["chars"] += len(line.get("content", ""))
+        speaker_stats[speaker]["last"] = i
+
+    print(f"\n说话人统计:")
+    for s, stats in sorted(speaker_stats.items(), key=lambda x: -x[1]["lines"]):
+        print(f"  {s}: {stats['lines']}行, {stats['chars']}字, 行范围[{stats['first']}-{stats['last']}]")
+
+    # 1.2 识别老师（发言最多的人）
+    teacher = max(speaker_stats, key=lambda s: speaker_stats[s]["lines"])
+    print(f"\n识别老师: {teacher} ({speaker_stats[teacher]['lines']}行)")
+
+    # 1.3 过滤杂音（< 20句的说话人）+ 排除老师
+    students = []
+    noise_speakers = []
+    for s, stats in speaker_stats.items():
+        if s == teacher:
+            continue
+        if stats["lines"] < 20:
+            noise_speakers.append(f"{s} ({stats['lines']}句)")
+            continue
+        students.append(s)
+
+    if noise_speakers:
+        print(f"\n杂音过滤: {', '.join(noise_speakers)}")
+    print(f"\n有效学员: {students}")
+
+    # 1.4 按时间范围截取：每个学员从第一句到最后一句，中间所有行全部截取
+    segments = []
+    for speaker in students:
+        stats = speaker_stats[speaker]
+        first_idx = stats["first"]
+        last_idx = stats["last"]
+        seg = lines[first_idx:last_idx + 1]
+
+        # 统计该学员在此范围内的发言
+        student_lines = [l for l in seg if l.get("speaker", "") == speaker]
+        student_chars = sum(len(l.get("content", "")) for l in student_lines)
+
+        segments.append((speaker, seg))
+        print(f"  {speaker}: 截取行[{first_idx}-{last_idx}], 共{len(seg)}行, 学员{len(student_lines)}句/{student_chars}字")
+
+    # 1.5 过滤：短段落 + 当事人占比过低（非主讲述人）
+    final_segments = []
+    filtered = []
+    for speaker, seg in segments:
+        student_lines = [l for l in seg if l.get("speaker", "") == speaker]
+        if len(student_lines) < 15:
+            filtered.append(f"{speaker} ({len(student_lines)}句<15)")
+            continue
+        student_chars = sum(len(l.get("content", "")) for l in student_lines)
+        if student_chars < 80:
+            filtered.append(f"{speaker} ({student_chars}字<80)")
+            continue
+        # 当事人说话占比 < 1/10 → 非主讲述人，丢弃
+        ratio = len(student_lines) / len(seg) if seg else 0
+        if ratio < 0.1:
+            filtered.append(f"{speaker} (占比{ratio:.1%}<10%)")
+            continue
+        final_segments.append((speaker, seg))
+
+    if filtered:
+        print(f"\n过滤: {', '.join(filtered)}")
+
+    print(f"\n最终: {len(final_segments)} 个学员段落")
+    for i, (speaker, seg) in enumerate(final_segments):
+        student_lines = [l for l in seg if l.get("speaker", "") == speaker]
+        student_chars = sum(len(l.get("content", "")) for l in student_lines)
+        ratio = len(student_lines) / len(seg) if seg else 0
+        print(f"  段落{i}: {speaker}, 截取{len(seg)}行, 学员{len(student_lines)}句/{student_chars}字, 占比{ratio:.1%}")
+
+    # 保存结果
+    if save:
+        step1_dir = os.path.join(output_dir, "step1_预处理")
+        # 保存每个学员段落
+        for i, (speaker, seg) in enumerate(final_segments):
+            text = _segment_to_text(seg)
+            safe_name = speaker.replace("/", "_")
+            _save_text(os.path.join(step1_dir, f"segment_{i}_{safe_name}.txt"), text)
+            print(f"  保存: segment_{i}_{safe_name}.txt ({len(text)}字)")
+        # 保存统计摘要
+        summary = {
+            "total_lines": len(lines),
+            "teacher": teacher,
+            "noise_filtered": noise_speakers,
+            "final_filtered": filtered,
+            "segments": [
+                {
+                    "speaker": s,
+                    "total_lines": len(seg),
+                    "student_lines": len([l for l in seg if l.get("speaker", "") == s]),
+                    "student_chars": sum(len(l.get("content", "")) for l in seg if l.get("speaker", "") == s),
+                    "ratio": round(len([l for l in seg if l.get("speaker", "") == s]) / len(seg), 3) if seg else 0,
+                }
+                for s, seg in final_segments
+            ]
+        }
+        _save_json(os.path.join(step1_dir, "summary.json"), summary)
+        print(f"  保存: summary.json")
+
+    return final_segments, teacher
+
+
+def _get_main_speaker(student_lines):
+    if not student_lines:
+        return "unknown"
+    counts = {}
+    for l in student_lines:
+        s = l.get("speaker", "")
+        counts[s] = counts.get(s, 0) + 1
+    return max(counts, key=counts.get)
+
+
+def _segment_to_text(seg):
+    parts = []
+    for line in seg:
+        speaker = line.get("speaker", "")
+        content = line.get("content", "")
+        if speaker:
+            parts.append(f"{speaker}: {content}")
+        else:
+            parts.append(content)
+    return "\n".join(parts)
+
+
+# ============================================================
+# 第二步：判断是否包含故事 + 提取事实（短段落合并处理）
+# ============================================================
+async def step2_identify(segments, output_dir, save=True):
+    """短段落直接提取事实，长段落先判断是否有故事再提取"""
+    print("\n" + "=" * 60)
+    print("第二步：判断故事 + 提取事实")
+    print("=" * 60)
+
+    SHORT_THRESHOLD = 4000  # 短段落阈值（字数）
+    all_facts = []
+    step2_results = []  # 记录每个段落的处理结果
+    step2_t0 = time.time()
+
+    # 构建所有段落的异步任务，并行执行
+    async def process_segment(i, speaker, seg):
+        text = _segment_to_text(seg)
+        safe_name = speaker.replace("/", "_")
+        prefix = f"seg{i}_{safe_name}_"
+        print(f"\n--- 段落 {i}: {speaker} ({len(text)} 字) ---")
+
+        result_record = {
+            "index": i,
+            "speaker": speaker,
+            "text_length": len(text),
+            "is_short": len(text) <= SHORT_THRESHOLD,
+        }
+
+        if len(text) <= SHORT_THRESHOLD:
+            print(f"  短段落，跳过判断，直接提取事实")
+            facts = await _extract_facts_from_chunk(text, "", output_dir, prefix)
+            if facts:
+                hints = facts.get("story_hints", "")
+                if not hints:
+                    hints = f"{speaker}的故事"
+                print(f"  提取完成: hints={hints}")
+                result_record["status"] = "extracted"
+                result_record["hints"] = hints
+                return (speaker, seg, hints, [facts]), result_record
+            else:
+                print(f"  提取失败，跳过")
+                result_record["status"] = "extract_failed"
+                return None, result_record
+        else:
+            is_story, hints = await _identify_story(text, output_dir, prefix)
+            print(f"  判断结果: is_story={is_story}, hints={hints}")
+            result_record["is_story"] = is_story
+            result_record["hints"] = hints
+
+            if is_story:
+                story_segments_for_step3 = [(speaker, seg, hints)]
+                step3_result = await step3_extract_facts(story_segments_for_step3, output_dir, save)
+                result_record["status"] = "has_story"
+                return step3_result, result_record
+            else:
+                print(f"  无故事，跳过")
+                result_record["status"] = "no_story"
+                return None, result_record
+
+    tasks = [process_segment(i, speaker, seg) for i, (speaker, seg) in enumerate(segments)]
+    results = await asyncio.gather(*tasks)
+
+    step2_elapsed = round(time.time() - step2_t0, 2)
+    print(f"\n⏱ 第二步总耗时: {step2_elapsed}s")
+
+    # 收集结果
+    for i, (r, record) in enumerate(results):
+        if r is not None:
+            if isinstance(r, list):
+                all_facts.extend(r)
+            else:
+                all_facts.append(r)
+        step2_results.append(record)
+
+    print(f"\n共 {len(all_facts)} 个故事段落完成事实提取")
+
+    # 保存结果
+    if save:
+        step2_dir = os.path.join(output_dir, "step2_判断提取")
+        _save_json(os.path.join(step2_dir, "results.json"), step2_results)
+        print(f"  保存: results.json")
+
+    return all_facts
+
+
+async def _identify_story(text, output_dir=None, log_prefix="", do_save=True):
+    prompt = """判断这段对话是否包含学员的个人故事。直接输出JSON，不要分析。
+
+个人故事特征（满足至少2条）：学员讲述具体经历、包含时间地点人物细节、表达深层情感、老师深入引导疗愈、揭示心理或家庭问题。
+
+非故事内容：纯课堂讲解、简单问答寒暄、课堂管理、纯疗愈引导词。
+
+直接输出JSON：
+{"is_story": true或false, "confidence": 0.0到1.0, "story_hints": "一句话描述故事主题和讲述者"}
+
+对话内容：
+```
+""" + text[:4000] + """
+```"""
+
+    try:
+        detail = await llm_client.chat_detail([{"role": "user", "content": prompt}], temperature=0.2)
+        api_log = {
+            "type": "identify_story",
+            "prompt": prompt,
+            "content": detail.get("content", ""),
+            "reasoning_content": detail.get("reasoning_content", ""),
+            "finish_reason": detail.get("finish_reason", ""),
+            "usage": detail.get("usage", {}),
+            "elapsed": detail.get("elapsed", 0),
+        }
+        print(f"  {log_prefix}API耗时: {api_log['elapsed']}s, finish_reason={api_log['finish_reason']}, "
+              f"tokens: {api_log['usage'].get('total_tokens', '?')} "
+              f"(reasoning={api_log['usage'].get('reasoning_tokens', 0)})")
+        if output_dir and do_save:
+            _save_json(os.path.join(output_dir, "api_logs", f"{log_prefix}identify.json"), api_log)
+        result = _parse_json(detail["content"])
+        if result:
+            return bool(result.get("is_story", False)) and result.get("confidence", 0) >= 0.5, result.get("story_hints", "")
+    except Exception as e:
+        print(f"  {log_prefix}错误: {e}")
+    return False, ""
+
+
+# ============================================================
+# 第三步：切片提取事实（Map）
+# ============================================================
+async def step3_extract_facts(story_segments, output_dir, save=True):
+    print("\n" + "=" * 60)
+    print("第三步：切片提取事实（Map）")
+    print("=" * 60)
+
+    all_facts = []
+    step3_t0 = time.time()
+    for i, (speaker, seg, hints) in enumerate(story_segments):
+        text = _segment_to_text(seg)
+        safe_name = speaker.replace("/", "_")
+        print(f"\n--- 故事 {i}: {speaker} - {hints[:50]}... ({len(text)} 字) ---")
+
+        chunks = _chunk_text_by_paragraphs(text, max_chars=3000, overlap_paragraphs=5)
+        print(f"  切成 {len(chunks)} 块")
+
+        # 保存切片信息
+        chunk_records = []
+        for j, chunk in enumerate(chunks):
+            chunk_records.append({"chunk_index": j, "chars": len(chunk), "preview": chunk[:100] + "..."})
+
+        tasks = [_extract_facts_from_chunk(chunk, hints, output_dir, f"story{i}_{safe_name}_chunk{j}_") for j, chunk in enumerate(chunks)]
+        results = await asyncio.gather(*tasks, return_exceptions=True)
+
+        facts = []
+        for j, r in enumerate(results):
+            if isinstance(r, Exception):
+                print(f"  块{j} 错误: {r}")
+                chunk_records[j]["status"] = "error"
+                chunk_records[j]["error"] = str(r)
+            else:
+                print(f"  块{j}: {json.dumps(r, ensure_ascii=False)[:200]}")
+                facts.append(r)
+                chunk_records[j]["status"] = "ok"
+                chunk_records[j]["facts"] = r
+
+        all_facts.append((speaker, seg, hints, facts))
+
+        # 保存切片和提取结果
+        if save:
+            step3_dir = os.path.join(output_dir, "step3_切片提取", f"story_{i}_{safe_name}")
+            for j, chunk in enumerate(chunks):
+                _save_text(os.path.join(step3_dir, f"chunk_{j}.txt"), chunk)
+            _save_json(os.path.join(step3_dir, "chunks_info.json"), chunk_records)
+            print(f"  保存: {len(chunks)} 个切片文件 + chunks_info.json")
+
+    step3_elapsed = round(time.time() - step3_t0, 2)
+    print(f"\n⏱ 第三步总耗时: {step3_elapsed}s")
+
+    return all_facts
+
+
+def _chunk_text_by_paragraphs(text, max_chars=3000, overlap_paragraphs=5):
+    """按自然段切分，在 max_chars 附近找段落结尾切，重叠 overlap_paragraphs 个自然段"""
+    paragraphs = [p for p in text.split("\n") if p.strip()]
+    if not paragraphs:
+        return [text]
+    if len(text) <= max_chars:
+        return [text]
+
+    chunks = []
+    i = 0
+    while i < len(paragraphs):
+        # 从当前位置开始，累加段落直到超过 max_chars
+        current = []
+        current_len = 0
+        j = i
+        while j < len(paragraphs):
+            para = paragraphs[j]
+            # 超过 max_chars 且已有内容 → 在这个段落前切断（段落结尾切）
+            if current_len + len(para) > max_chars and current:
+                break
+            current.append(para)
+            current_len += len(para)
+            j += 1
+
+        chunks.append("\n".join(current))
+
+        # 到末尾了就结束
+        if j >= len(paragraphs):
+            break
+
+        # 下一次起点：当前切到的位置 - 重叠段落数
+        next_i = j - overlap_paragraphs
+        # 防止死循环：确保至少前进1个段落
+        if next_i <= i:
+            next_i = i + 1
+        i = next_i
+
+    return chunks
+
+
+async def _extract_facts_from_chunk(text, hints, output_dir=None, log_prefix="", do_save=True):
+    prompt = f"""从这段对话中提取关键信息。只输出JSON。
+
+故事主题：{hints}
+
+提取以下内容：
+- events: 发生的事件（尽量详细，保留具体细节）
+- emotions: 表达的情感
+- quotes: 讲述者的原话（保留原话，不要改写，尽量多提取；完全相同的重复语句只保留一条）
+- key_people: 提到的人物
+
+{{"events":["事件1","事件2"],"emotions":["情感1","情感2"],"quotes":["原话1","原话2","原话3"],"key_people":["人物1"]}}
+
+对话内容：
+```
+{text}
+```"""
+
+    detail = await llm_client.chat_detail([{"role": "user", "content": prompt}], temperature=0.1)
+    api_log = {
+        "type": "extract_facts",
+        "prompt": prompt,
+        "content": detail.get("content", ""),
+        "reasoning_content": detail.get("reasoning_content", ""),
+        "finish_reason": detail.get("finish_reason", ""),
+        "usage": detail.get("usage", {}),
+        "elapsed": detail.get("elapsed", 0),
+    }
+    print(f"  {log_prefix}API耗时: {api_log['elapsed']}s, finish_reason={api_log['finish_reason']}, "
+          f"tokens: {api_log['usage'].get('total_tokens', '?')} "
+          f"(reasoning={api_log['usage'].get('reasoning_tokens', 0)})")
+    if output_dir and do_save:
+        _save_json(os.path.join(output_dir, "api_logs", f"{log_prefix}extract.json"), api_log)
+    result = _parse_json(detail["content"])
+    return result
+
+
+# ============================================================
+# 模糊去重工具函数
+# ============================================================
+def _fuzzy_dedup_quotes(quotes: list[str]) -> list[str]:
+    """
+    模糊去重：对高度相似的语句只保留第一条。
+    策略：提取每条语句的核心部分（去掉常见前缀后缀），相似度超过阈值的视为重复。
+    """
+    import re
+
+    def normalize(s: str) -> str:
+        s = s.strip()
+        if s.startswith("「") and s.endswith("」"):
+            s = s[1:-1]
+        for prefix in ["嗯", "啊", "呃", "就是", "然后", "那个", "所以"]:
+            if s.startswith(prefix):
+                s = s[len(prefix):].strip()
+        for suffix in ["啊", "吧", "呢", "嘛", "哦", "呀"]:
+            if s.endswith(suffix) and len(s) > 4:
+                s = s[:-1]
+        s = re.sub(r'(.)\1{2,}', r'\1', s)
+        return s
+
+    def similarity(a: str, b: str) -> float:
+        if not a or not b:
+            return 0.0
+        a_bigrams = set(a[i:i+2] for i in range(len(a)-1))
+        b_bigrams = set(b[i:i+2] for i in range(len(b)-1))
+        if not a_bigrams or not b_bigrams:
+            return 0.0
+        intersection = a_bigrams & b_bigrams
+        union = a_bigrams | b_bigrams
+        return len(intersection) / len(union)
+
+    kept = []
+    kept_normalized = []
+    for q in quotes:
+        nq = normalize(q)
+        is_dup = False
+        for kn in kept_normalized:
+            threshold = 0.8 if len(nq) < 15 else 0.7
+            if similarity(nq, kn) >= threshold:
+                is_dup = True
+                break
+        if not is_dup:
+            kept.append(q)
+            kept_normalized.append(nq)
+
+    removed = len(quotes) - len(kept)
+    if removed > 0:
+        print(f"    模糊去重: {len(quotes)} -> {len(kept)} (去除 {removed} 条相似重复)")
+
+    return kept
+
+
+# ============================================================
+# 第四步：重组（Reduce）
+# ============================================================
+def step4_merge(all_facts, output_dir, save=True):
+    print("\n" + "=" * 60)
+    print("第四步：重组（Reduce）")
+    print("=" * 60)
+
+    merged = []
+    for i, (speaker, seg, hints, facts_list) in enumerate(all_facts):
+        print(f"\n--- 故事 {i}: {speaker} ---")
+
+        all_events, all_emotions, all_quotes, all_people = [], [], [], []
+        for facts in facts_list:
+            if isinstance(facts, dict):
+                all_events.extend(facts.get("events", []))
+                all_emotions.extend(facts.get("emotions", []))
+                all_quotes.extend(facts.get("quotes", []))
+                all_people.extend(facts.get("key_people", []))
+
+        all_emotions = list(dict.fromkeys(all_emotions))
+        all_people = list(dict.fromkeys(all_people))
+        seen = set()
+        unique_quotes = [q for q in all_quotes if q not in seen and not seen.add(q)]
+        # 模糊去重：去除相似重复语句（如多次释放语句只保留一条）
+        unique_quotes = _fuzzy_dedup_quotes(unique_quotes)
+
+        material = {
+            "events": all_events,
+            "emotions": all_emotions,
+            "quotes": unique_quotes,
+            "key_people": all_people,
+            "story_hints": hints,
+        }
+
+        print(f"  事件: {len(all_events)}, 情感: {len(all_emotions)}, 原话: {len(unique_quotes)}, 人物: {len(all_people)}")
+        print(f"  素材: {json.dumps(material, ensure_ascii=False)[:500]}")
+
+        merged.append((speaker, seg, material))
+
+    # 保存结果
+    if save:
+        step4_dir = os.path.join(output_dir, "step4_重组")
+        for i, (speaker, seg, material) in enumerate(merged):
+            safe_name = speaker.replace("/", "_")
+            _save_json(os.path.join(step4_dir, f"material_{i}_{safe_name}.json"), material)
+            print(f"  保存: material_{i}_{safe_name}.json")
+
+    return merged
+
+
+# ============================================================
+# 第五步：生成最终故事
+# ============================================================
+async def step5_generate(merged, output_dir, save=True):
+    print("\n" + "=" * 60)
+    print("第五步：生成最终故事（并行）")
+    print("=" * 60)
+
+    step5_t0 = time.time()
+
+    async def generate_one(i, speaker, seg, material):
+        print(f"\n--- 生成故事 {i}: {speaker} ---")
+
+        prompt = f"""基于素材生成第三人称故事。只输出JSON，不要其他内容。
+
+素材：
+主题：{material['story_hints']}
+事件：{json.dumps(material['events'], ensure_ascii=False)}
+情感：{json.dumps(material['emotions'], ensure_ascii=False)}
+原话：{json.dumps(material['quotes'], ensure_ascii=False)}
+人物：{json.dumps(material['key_people'], ensure_ascii=False)}
+
+写作手法（非常重要）：
+
+【双时空结构】
+故事有两个时空层：
+- "当下时空"：学员在课堂上与卢慧老师的对话——这是故事的框架，用对话体呈现，保留师生互动的完整脉络
+- "过往时空"：学员讲述的过去经历——这是故事的核心，用"场景重建"手法呈现，让读者穿越到那个时空
+
+【场景重建手法（用于"过往的故事"部分）】
+当学员讲述过去的经历时，不要写成"她说她小时候…"这种转述体，而是直接把读者拉进那个时空：
+- 直接切入场景，用过去时态叙述。比如不要写"文晔回忆说她的儿子15岁去了美国"，而要写"十五岁那年，文晔的儿子独自登上了飞往美国的航班"
+- 用五感描写构建画面：她看到了什么、听到了什么、身体感受到了什么（比如"手心出汗"、"脚发凉"这些身体反应要融入场景中）
+- 用环境细节暗示情绪：不要直接说"她很痛苦"，而是通过场景让读者感受到痛苦
+- 关键情感爆发处用直接引语（「」包裹原话），其他地方用间接引语和叙述自然融合
+- 场景之间要有过渡，比如"然而，命运的转折发生在前年冬天——"
+
+【对话体手法（用于"当下的困境"和"转变"部分）】
+- 保留师生对话的现场感，写出谁说了什么、对方如何回应
+- 在对话中穿插学员的内心活动、身体反应、情绪变化
+- 老师的引导语要完整保留，这是疗愈过程的关键
+
+【通用要求】
+- 每个事件都要展开写，写出具体过程和细节，不要一句话概括
+- 事件之间要有过渡和因果逻辑，不能跳跃
+- 情感变化要写出过程：从什么情感转变到什么情感，中间经历了什么
+- 不编造素材中没有的内容，但可以根据素材合理补充场景细节（如环境、氛围）
+- 每个部分充分展开，不要限制长度，越详细越好
+- 原文金句数组中的每条金句必须用英文双引号包裹，如 ["金句1", "金句2"]，不要用「」替代双引号
+
+禁止事项：
+- 禁止用一句话概括多个事件（如"她倾诉了挣扎"、"她经历了疗愈过程"）
+- 禁止省略素材中的具体细节
+- 禁止省略老师的引导和回应
+- 禁止事件之间跳跃，要保持时间顺序和因果逻辑
+- 禁止过度总结或抽象化描述
+- 禁止在"过往的故事"部分全程使用转述体（"她说…她回忆…"），必须用场景重建手法让读者身临其境
+- 禁止为了简洁而牺牲内容的丰富性
+
+输出格式：{{"title":"标题","summary":"80字摘要","tags":["标签"],"speaker_nickname":"昵称","content":{{"讲述者":"...","当下的困境":"...","过往的故事":"...","转变":"...","原文金句":["原话1","原话2","原话3"]}}}}"""
+
+        try:
+            detail = await llm_client.chat_detail([{"role": "user", "content": prompt}], temperature=0.7, max_tokens=16384)
+            raw = detail["content"]
+            api_log = {
+                "type": "generate_story",
+                "speaker": speaker,
+                "prompt": prompt,
+                "content": raw,
+                "reasoning_content": detail.get("reasoning_content", ""),
+                "finish_reason": detail.get("finish_reason", ""),
+                "usage": detail.get("usage", {}),
+                "elapsed": detail.get("elapsed", 0),
+            }
+            print(f"  API耗时: {api_log['elapsed']}s, finish_reason={api_log['finish_reason']}, "
+                  f"tokens: {api_log['usage'].get('total_tokens', '?')} "
+                  f"(reasoning={api_log['usage'].get('reasoning_tokens', 0)})")
+            print(f"  原始输出长度: {len(raw)} 字")
+            if save:
+                step5_dir = os.path.join(output_dir, "step5_生成故事")
+                os.makedirs(step5_dir, exist_ok=True)
+                _save_text(os.path.join(step5_dir, f"raw_{i}_{speaker}.txt"), raw)
+                _save_json(os.path.join(output_dir, "api_logs", f"story{i}_{speaker}_generate.json"), api_log)
+            result = _parse_json(raw)
+        except Exception as e:
+            print(f"  生成失败: {e}")
+            result = None
+
+        if result:
+            print(f"  标题: {result.get('title', '')}")
+            print(f"  摘要: {result.get('summary', '')[:100]}")
+            content = result.get('content', {})
+            if isinstance(content, dict):
+                for key, val in content.items():
+                    print(f"  {key}: {str(val)[:80]}...")
+            print(f"\n  完整JSON:")
+            print(json.dumps(result, ensure_ascii=False, indent=2))
+        else:
+            print("  生成失败！")
+
+        return result
+
+    tasks = [generate_one(i, speaker, seg, material) for i, (speaker, seg, material) in enumerate(merged)]
+    results = await asyncio.gather(*tasks)
+    stories = [r for r in results if r is not None]
+
+    step5_elapsed = round(time.time() - step5_t0, 2)
+    print(f"\n⏱ 第五步总耗时: {step5_elapsed}s")
+
+    # 保存结果
+    if save:
+        step5_dir = os.path.join(output_dir, "step5_生成故事")
+        for i, story in enumerate(stories):
+            safe_name = story.get("speaker_nickname", f"story_{i}").replace("/", "_")
+            _save_json(os.path.join(step5_dir, f"story_{i}_{safe_name}.json"), story)
+            print(f"  保存: story_{i}_{safe_name}.json")
+
+    return stories
+
+
+# ============================================================
+# 主函数
+# ============================================================
+async def main():
+    if len(sys.argv) < 2:
+        print("用法: python test_pipeline.py <docx文件路径> [步骤编号]")
+        print("  步骤编号: 1=预处理, 2=判断故事, 3=切片提取, 4=重组, 5=生成故事, all=全部")
+        sys.exit(1)
+
+    file_path = sys.argv[1]
+    step = sys.argv[2] if len(sys.argv) > 2 else "all"
+
+    if not os.path.exists(file_path):
+        print(f"文件不存在: {file_path}")
+        sys.exit(1)
+
+    # 根据文件名创建输出目录
+    basename = os.path.splitext(os.path.basename(file_path))[0]
+    output_dir = os.path.join(TEST_DIR, basename)
+    os.makedirs(output_dir, exist_ok=True)
+
+    print(f"文件: {file_path}")
+    print(f"输出目录: {output_dir}")
+
+    total_t0 = time.time()
+
+    lines = parse_transcript(file_path)
+    print(f"解析到 {len(lines)} 行")
+
+    # 如果只跑第5步，从 step4 产物恢复
+    if step == "5":
+        step4_dir = os.path.join(output_dir, "step4_重组")
+        if os.path.exists(step4_dir):
+            print("从 step4 产物恢复...")
+            merged = []
+            for f in sorted(os.listdir(step4_dir)):
+                if f.endswith(".json"):
+                    with open(os.path.join(step4_dir, f), "r", encoding="utf-8") as fp:
+                        material = json.load(fp)
+                    parts = f.replace("material_", "").replace(".json", "").split("_", 1)
+                    speaker = parts[1] if len(parts) > 1 else f
+                    merged.append((speaker, [], material))
+            print(f"恢复到 {len(merged)} 个故事")
+            stories = await step5_generate(merged, output_dir)
+
+            total_elapsed = round(time.time() - total_t0, 2)
+            print(f"\n{'=' * 60}")
+            print(f"✅ 全流程完成！总耗时: {total_elapsed}s")
+            print(f"   生成故事: {len(stories)} 个")
+            print(f"{'=' * 60}")
+            return
+        else:
+            print("step4 产物不存在，请先跑前4步")
+            return
+
+    segments, teacher = step1_preprocess(lines, output_dir)
+    if step == "1":
+        return
+
+    all_facts = await step2_identify(segments, output_dir)
+    if step == "2":
+        return
+
+    if not all_facts:
+        print("没有发现故事段落，结束。")
+        return
+
+    merged = step4_merge(all_facts, output_dir)
+    if step == "4":
+        return
+
+    stories = await step5_generate(merged, output_dir)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())