DocuMind — AI PDF Chatbot

Upload PDFs and chat with them using AI. Get cited, grounded answers with page references.

Live Demo: https://documind-frontend-static.onrender.com
Backend API: https://documind-backend-free.onrender.com/docs

---

What it does

• Upload one or more PDF documents
• Ask natural language questions about their contents
• Get accurate answers with inline citations (filename + page number)
• Chat across multiple documents at once
• Conversations are persisted — pick up where you left off

---

Tech Stack

Layer	Technology
Frontend	React 18, Vite, custom CSS (glassmorphism)
Backend	FastAPI (Python 3.11)
LLM	Groq — llama-3.3-70b-versatile
Embeddings	Google Gemini — gemini-embedding-2
Vector Search	Supabase pgvector (cosine similarity)
Keyword Search	BM25 (rank-bm25) — hybrid retrieval with RRF fusion
Database	Supabase Postgres
File Storage	Supabase Storage (private bucket)
Auth	JWT + Google OAuth 2.0
Deployment	Render (frontend static site + backend Docker)

---

Architecture

User → React Frontend (Render Static)
          ↓ HTTPS
     FastAPI Backend (Render Docker)
          ↓
     ┌────────────────────────────────┐
     │  PDF Upload → Supabase Storage │
     │  Parse (PyMuPDF + Tesseract)   │
     │  Chunk → Embed (Gemini)        │
     │  Store → Supabase pgvector     │
     └────────────────────────────────┘
          ↓  Query time
     Hybrid Retrieval (pgvector + BM25 + RRF)
          ↓
     Groq LLM (streaming SSE)
          ↓
     Cited answer → User

---

Retrieval Pipeline

1. Dense vector search — Gemini embeddings + pgvector cosine similarity
2. Sparse BM25 search — keyword matching over all chunks
3. RRF fusion — Reciprocal Rank Fusion combines both ranked lists
4. Reranking — term overlap + section heading boost + exact phrase bonus
5. Summary injection — for summary questions, document summary is prepended to context

---

Features

• Hybrid RAG — vector + BM25 + RRF for higher retrieval accuracy
• Streaming responses — SSE streaming via Groq, no wait for full answer
• Source citations — every answer cites filename and page number
• Multi-document chat — ask across several PDFs in one conversation
• Processing queue — durable background worker with heartbeat, retry, and recovery
• Restart-safe — files in Supabase Storage, vectors in pgvector, all persisted across redeploys
• Google OAuth — sign in with Google or email/password
• Responsive dark UI — glassmorphism design, skeleton loading, no flash states

---

Running Locally

Backend

cd backend
cp ../.env.example .env   # fill in your keys
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Frontend

cd frontend
npm install
npm run dev

Frontend runs at http://localhost:5173, proxies /api to http://localhost:8000.

---

Environment Variables

Variable	Description
DATABASE_URL	Supabase Postgres Session Pooler URI
SUPABASE_URL	Supabase project URL
SUPABASE_SERVICE_ROLE_KEY	Supabase service role key (backend only)
SUPABASE_STORAGE_BUCKET	Storage bucket name (pdf-documents)
GROQ_API_KEY	Groq API key
GEMINI_API_KEY	Google Gemini API key
GOOGLE_CLIENT_ID	Google OAuth client ID
GOOGLE_CLIENT_SECRET	Google OAuth client secret
JWT_SECRET_KEY	Secret for signing JWTs
FRONTEND_URL	Frontend origin (for CORS + OAuth redirect)

---

Deployment

Both services deploy automatically on push to master via render.yaml.

• Frontend — Render Static Site, Vite build, SPA rewrite rules
• Backend — Render Docker (python:3.11-slim + Tesseract OCR)

Note: Render free tier spins down after 15 min inactivity. First load after idle may take ~30–60 seconds while the backend wakes up. The UI shows a "waking up" message during this time.

---

Project Structure

.
├── backend/
│   ├── app/
│   │   ├── api/          # FastAPI routers (auth, chat, documents)
│   │   ├── core/         # Config, database, security
│   │   ├── models/       # SQLAlchemy ORM models
│   │   └── services/     # PDF parser, chunker, embedder, retriever, LLM, storage
│   ├── Dockerfile
│   └── requirements.txt
├── frontend/
│   ├── src/
│   │   ├── components/   # React UI components
│   │   ├── hooks/        # useAuth, useDocuments, useChat
│   │   └── services/     # API client (axios + fetch for SSE)
│   ├── Dockerfile
│   └── vite.config.js
└── render.yaml           # Render deployment config

---

Built by Kanish Mehan