DocRAG — Chat with your PDFs

DocRAG is a full-stack Retrieval-Augmented Generation (RAG) application that lets you upload multiple PDF documents and have a natural language conversation across all of them simultaneously.

---

Features

• Multi-PDF support — Upload multiple PDFs and query across all of them in a single chat
• Fast background ingestion — Large PDFs are processed in background threads with parallel embedding batches; real-time progress bars show indexing status
• Vector search — Qdrant (in-memory) for accurate semantic retrieval
• Conversation history — Full multi-turn chat with context passed to the LLM
• User authentication — JWT-based signup/login
• Re-index on demand — One-click re-index button re-ingests saved PDFs after a server restart
• Groq LLM — Uses llama-3.3-70b-versatile via Groq for fast, free-tier-friendly responses
• Google Embeddings — gemini-embedding-001 (3072-dim) for high-quality semantic search

---

Tech Stack

Layer	Technology
Frontend	React 18, Vite, Tailwind CSS, Axios
Backend	Python, Flask, Flask-CORS
LLM	Groq API (llama-3.3-70b-versatile)
Embeddings	Google Generative AI (gemini-embedding-001)
Vector Store	Qdrant (in-memory)
Auth	JWT (PyJWT)
PDF Parsing	pypdf, LangChain Community

---

Project Structure

├── app/
│   ├── api/
│   │   ├── app.py              # Flask routes (ingest, query, auth, documents)
│   │   ├── auth.py             # Signup/login with JWT
│   │   ├── chat_storage.py     # In-memory conversation store
│   │   └── middleware.py       # JWT token verification
│   ├── config/
│   │   └── db.py               # Qdrant in-memory client + collection init
│   ├── generator/
│   │   └── llm.py              # Groq LLM answer generation
│   ├── ingestion/
│   │   ├── ingestion_pipeline.py  # Parallel batch ingestion with progress tracking
│   │   ├── loader.py           # PDF loader (pypdf)
│   │   ├── chunking.py         # Text splitting
│   │   ├── embedder.py         # Google embedding model
│   │   └── store_to_db.py      # Qdrant upsert with PointStruct
│   ├── retrival/
│   │   └── retrieve_query.py   # Vector search + chunk retrieval
│   └── main_pipeline.py        # Retrieve → Generate orchestration
├── frontend/
│   ├── src/
│   │   ├── App.jsx             # Main React UI
│   │   ├── api.js              # Axios client (proxied to backend)
│   │   └── main.jsx            # React entry point
│   ├── vite.config.js          # Vite + proxy config (port 5000)
│   └── package.json
├── run_server.py               # Flask server entry point (port 8000)
├── requirements.txt
└── README.md

---

Getting Started

Prerequisites

• Python 3.12+
• Node.js 20+
• A Google AI API key (for embeddings — free)
• A Groq API key (for LLM chat — free tier)

1. Clone the repository

git clone https://github.com/Nirvanjha2004/DocRAG.git
cd DocRAG

2. Install backend dependencies

pip install -r requirements.txt

3. Install frontend dependencies

cd frontend && npm install

4. Set environment variables

Create a .env file in the project root or export in your shell:

GOOGLE_API_KEY=your_google_ai_api_key
GROQ_API_KEY=your_groq_api_key

5. Run the application

Backend (port 8000):

python run_server.py

Frontend (port 5000):

cd frontend && npm run dev

Open http://localhost:5000 in your browser.

---

How It Works

1. Upload — Select one or more PDF files via the sidebar. Files are saved to data/uploads/<username>/
2. Ingestion — Each PDF is loaded, chunked into 800-char pieces, and embedded in parallel background threads. A real-time progress bar shows status per file
3. Search — On a query, your question is embedded and used to find the most semantically similar chunks from all indexed documents
4. Answer — Retrieved chunks + conversation history are sent to Groq's LLaMA model which generates a context-aware response

---

API Endpoints

Method	Endpoint	Description
GET	/api/health	Health check
POST	/api/auth/signup	Register new user
POST	/api/auth/login	Login, returns JWT
GET	/api/documents	List uploaded PDFs
POST	/api/ingest	Upload + start background indexing (multi-file)
GET	/api/ingest/status/<job_id>	Poll indexing progress
POST	/api/reingest	Re-index an already-uploaded PDF
POST	/api/conversations	Create a new conversation
GET	/api/conversations	List conversations
GET	/api/conversations/<id>/messages	Get messages in a conversation
POST	/api/query	Ask a question (RAG)

---

Notes

• In-memory vector store — Qdrant runs in-memory. Indexed documents are lost on server restart. Use the Re-index (↺) button in the sidebar to re-ingest your uploaded PDFs
• Large PDFs — Pages are batched (40 pages/batch) with up to 3 parallel embedding calls. A 500-page PDF typically takes 2–5 minutes depending on Google Embedding API rate limits

---

License

MIT