It is intentionally separated from the existing hy_qa_asset_backend/docs folder because the old implementation was an MVP for "大本营答疑资产后台系统", while the new requirement is a broader AI knowledge base product with login, permission control, chat sessions, admin management, Feishu real-time retrieval, and RAG.

Directory Layout

source_documents/: original PRD, SRS, API, database, RAG, prototype, plan, test, UAT, deployment, data dictionary, and permission matrix files.
extracted/: machine-extracted Markdown versions of the source documents for search and review.
development_records/: new project development notes, decisions, scope records, and future implementation logs.
reports/: assessment and design reports for this new version.

Current Goal

Build a new version based on the uploaded AI knowledge base requirement package:

User side: phone/SMS login, AI chat, sessions, history, streaming response, stop generation, Markdown rendering, and knowledge permission filtering.
Admin side: admin login, user management, admin management, knowledge base management, Prompt management, model management, chat records, dashboard, system config, and operation logs.
AI/RAG: Feishu knowledge base as the only knowledge source in phase one, real-time retrieval per question, permission-based knowledge filtering, Prompt assembly, model invocation, SSE streaming, and AI request logging.
Phase-one exclusions: file upload, OCR, image understanding, web search, local vector database sync, payment, course system, automatic authorization, multi-tenant, multi-Prompt, regenerate answer, and edit message.

Working Principles

Keep this version additive and isolated until the architecture decision is confirmed.
Do not put new planning records into the old hy_qa_asset_backend/docs directory.
Treat the uploaded documents as the source of truth for V2.
Record major implementation decisions in development_records/ before or alongside code changes.
Reuse old code only where the domain boundary still matches; do not force old "QA asset review" models into the new "AI knowledge base chat" model.