DocMind

RAG-powered document Q&A system. Upload PDF/TXT documents, get AI-generated summaries, and ask questions with SSE streaming responses grounded in your documents.

Overview

DocMind demonstrates practical implementation of Retrieval-Augmented Generation (RAG) — a pattern essential for enterprise AI applications. It processes uploaded documents, chunks them for efficient retrieval, and answers questions strictly based on document content.

Features

• Document upload: Drag-and-drop PDF/TXT (up to 10MB)
• AI summarization: Automatic document abstract via DeepSeek API
• Smart chunking: 500-character chunks with 100-character overlap for optimal retrieval
• SSE streaming Q&A: Real-time, grounded answers that cite document sources
• Keyword retrieval: Simplified BM25-based relevance matching
• Document management: List, view, and delete uploaded documents

Tech Stack

Layer	Technology
Frontend	React 19, Vite, Tailwind CSS, TypeScript
Backend	Express, multer (upload), pdf-parse
Database	SQLite (document metadata + chunk storage)
AI	DeepSeek API (SSE streaming)
Retrieval	Custom BM25-style keyword matching

API Reference

Method	Endpoint	Description
GET	/health	Health check
POST	/api/documents	Upload document (multipart)
GET	/api/documents	List all documents
GET	/api/documents/:id	Get document details
DELETE	/api/documents/:id	Remove document
POST	/api/chat	SSE streaming Q&A

Quick Start

git clone https://github.com/Suk-Builder/docmind.git
cd docmind
cp .env.example .env
# Add your DEEPSEEK_API_KEY
npm install
npm run build
npm start
# http://localhost:443 (or PORT in .env)

RAG Pipeline

PDF/TXT Upload
    → Text extraction (pdf-parse)
    → Chunking (500 chars, 100 overlap)
    → BM25 keyword indexing
    → SQLite storage

User Question
    → Keyword matching → Retrieve top-k chunks
    → DeepSeek API (context + question)
    → SSE stream response with source references

Why This Matters for Germany

RAG architecture is a core requirement for DiGA-compliant medical documentation systems and enterprise knowledge management in GDPR-regulated environments. DocMind demonstrates practical implementation of:

• Document ingestion and processing pipelines
• Vector/text hybrid retrieval
• Streaming AI responses with source attribution
• Data privacy (on-premise deployment, no external data leakage)

Deployment

# PM2
pm2 start ecosystem.config.js
pm2 save

# Docker (if available)
docker-compose up -d

Environment Variables

Variable	Default	Description
PORT	443	Server port
DEEPSEEK_API_KEY	—	DeepSeek API key
DB_PATH	./db/docmind.db	SQLite database path

About

Built by Ying Momo — Computer Science undergraduate, targeting Germany's AI and digital health market.

License

MIT

---

与Builder-System的关系

本项目属于Builder-System 域II — AI认知：AI文档问答系统。FastAPI+React 19，Builder-System文档知识库。

Builder-System（V4.3，104篇文本、35元概念）→ 了解更多