RAG Knowledge Base Q&A System

企业级 RAG 知识库问答系统 — 双模式部署，多模型切换，多格式文档解析，智能检索增强生成。

当前状态: ✅ 全链路跑通 (DeepSeek + Chroma + sentence-transformers)

Architecture

User → FastAPI → RAG Service → LLM (Ollama/DeepSeek/Qwen)
                    │
                    ├── Vector Retriever → Chroma / Milvus
                    ├── Reranker → Local CrossEncoder / DashScope
                    ├── Embedding → Sentence-Transformers / DashScope
                    └── Document Pipeline → PDF/DOCX/MD/TXT → Chunks → Vectors

Tech Stack

Layer	Lightweight (default)	Enterprise
LLM	Ollama / DeepSeek	Qwen-Max
Vector DB	ChromaDB	Milvus
Embedding	sentence-transformers	DashScope
Rerank	CrossEncoder	gte-rerank
Storage	Local filesystem	MinIO
Queue	Sync (in-process)	Celery + RabbitMQ
Cache	In-memory dict	Redis
Frontend	Streamlit	Vue 3

Quick Start

1. Install

pip install -r requirements.txt

2. Configure .env

# Pick your LLM
LLM_PROVIDER=deepseek       # or ollama, dashscope
DEEPSEEK_API_KEY=sk-xxx     # for DeepSeek

# HF mirror for China users
HF_ENDPOINT=https://hf-mirror.com

3. One-click demo (no server needed)

python demo.py

4. Start server

uvicorn src.main:app --host 0.0.0.0 --port 8000 --reload
# API docs: http://localhost:8000/docs
# Streamlit: streamlit run streamlit_app.py

5. Docker (single container)

docker-compose -f docker/docker-compose.light.yml up -d

API

Method	Path	Description
POST	/api/v1/auth/register	Register
POST	/api/v1/auth/login/access-token	Login
POST	/api/v1/knowledge-bases/	Create KB
POST	/api/v1/knowledge-bases/{id}/upload	Upload doc
POST	/api/v1/chat/	RAG query
GET	/api/v1/chat/sessions	History
POST	/api/v1/evaluations/	Evaluate
GET	/api/v1/health/	Health

Supported Formats

PDF / DOCX / Markdown / TXT / Excel / PowerPoint (enterprise)

Docker (Enterprise)

docker-compose -f docker/docker-compose.yml up -d

Features

• Dual-mode: Chroma + Ollama for dev, Milvus + Qwen-Max for prod
• Multi-LLM: Switch by changing one config line
• Hybrid search: BM25 (keyword) + Vector (semantic) + RRF fusion, CrossEncoder reranking
• Multi-hop: Auto question decomposition for complex queries
• Memory: Short-term conversation + long-term vector memory
• Evaluation: RAGAS evaluation with auto QA pair generation

Project Structure

ragPdfSystem/
├── src/
│   ├── main.py              # FastAPI entry
│   ├── settings.py          # Config (from .env)
│   ├── api/routers/         # 12 routers
│   ├── services/            # RAG, Evaluation, Memory
│   ├── database/            # SQL + Chroma/Milvus
│   ├── retrieval/           # Retriever + Reranker
│   ├── embedding/           # Multi-provider
│   ├── llm/                 # Multi-model
│   └── processors/          # PDF, DOCX parsers
├── streamlit_app.py         # Streamlit UI
├── frontend/                # Vue 3 UI
├── docker/                  # Docker
└── requirements.txt