RAG Research Assistant

A full-stack Retrieval-Augmented Generation web application that lets you upload documents (PDF, TXT, MD) and ask natural language questions about them. Answers are grounded strictly in your uploaded content — no hallucinations.

Deployed and publicly accessible. Zero cost.

---

Features

• 📤 Upload documents — drag-and-drop PDF, TXT, or Markdown files
• 🔍 Semantic search — vector similarity search using pgvector + sentence-transformers
• 🤖 Blazing-fast AI answers — powered by Groq API (LLaMA 3.3 70B)
• 📎 Source citations — see exactly which documents were used
• ⚡ Background ingestion — upload returns immediately, processing happens async
• 🗑️ Document management — view status, delete documents and their vectors
• 💬 Chat interface — sleek Q&A with message history
• ☁️ Fully cloud-hosted — accessible from anywhere

---

Tech Stack

Layer	Technology
Backend	Python 3.11 · FastAPI · SQLAlchemy · Gunicorn
Embeddings	sentence-transformers (all-MiniLM-L6-v2)
LLM	Groq API (LLaMA 3.3 70B Versatile)
Vector Search	PostgreSQL + pgvector (cosine similarity)
Document Parsing	LangChain community loaders
Frontend	React · Vite · Tailwind CSS · Axios
Database	Supabase PostgreSQL (metadata + vectors)
File Storage	Supabase Storage
Backend Hosting	Render / Railway
Frontend Hosting	Vercel

---

Architecture

┌─────────────┐     HTTPS      ┌──────────────────┐     REST      ┌─────────────┐
│   Vercel     │ ──────────────▶│  Render/Railway   │ ────────────▶│  Groq API   │
│  (React UI)  │                │  (FastAPI Server)  │              │  (LLM)      │
└─────────────┘                └──────────────────┘              └─────────────┘
                                       │
                                       │ SQL + pgvector
                                       ▼
                              ┌──────────────────┐
                              │    Supabase       │
                              │  PostgreSQL +     │
                              │  Storage          │
                              └──────────────────┘

---

Quick Start (Local Development)

Prerequisites

• Python 3.11+
• Node.js 18+
• A Groq API key (free)
• A Supabase project (free)

1. Set up Supabase

1. Create a new Supabase project
2. Go to SQL Editor and run: CREATE EXTENSION IF NOT EXISTS vector;
3. Create a Storage bucket called documents (set to public or use service key)
4. Copy your Project URL, Service Role Key, and Database URI

2. Configure environment

cd backend
cp .env.example .env
# Fill in your Groq API key and Supabase credentials

3. Start the backend

cd backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

4. Start the frontend

cd frontend
npm install
npm run dev

5. Open the app

• App: http://localhost:5173
• API Docs: http://localhost:8000/docs

---

Cloud Deployment

Backend → Render / Railway

1. Push your repo to GitHub
2. Create a new Web Service on Render
3. Set Root Directory to backend
4. Set Build Command: pip install -r requirements.txt
5. Set Start Command: gunicorn main:app -w 2 -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:$PORT
6. Add all env vars from .env.example

Frontend → Vercel

1. Import your repo to Vercel
2. Set Root Directory to frontend
3. Set Framework Preset to Vite
4. Add env var: VITE_API_BASE_URL = your Render backend URL

---

API Endpoints

Method	Endpoint	Description
POST	/api/documents/upload	Upload a document
GET	/api/documents	List all documents
GET	/api/documents/{id}	Get document status
DELETE	/api/documents/{id}	Delete a document
POST	/api/ask	Ask a question
POST	/api/ask/stream	Ask with streaming (SSE)
GET	/health	Health check

---

Project Structure

├── backend/
│   ├── main.py               # FastAPI app entry point
│   ├── config.py             # Central configuration (env vars)
│   ├── database.py           # SQLAlchemy + pgvector models
│   ├── ingest.py             # Document ingestion pipeline
│   ├── retriever.py          # pgvector similarity search
│   ├── llm.py                # Groq API integration
│   ├── routes/
│   │   ├── documents.py      # Upload, list, delete routes
│   │   └── query.py          # Q&A routes
│   ├── models/
│   │   └── schemas.py        # Pydantic schemas
│   └── utils/
│       └── file_handler.py   # Supabase Storage utilities
├── frontend/
│   ├── src/
│   │   ├── App.jsx           # Root layout
│   │   ├── components/       # React components
│   │   ├── hooks/            # Custom hooks
│   │   └── api/              # Axios client
│   └── vite.config.js        # Vite config
└── README.md

---

Future Enhancements

• Streaming responses with SSE (route already scaffolded)
• Conversation memory (pass last N messages to LLM)
• Hybrid search: BM25 keyword + vector search
• Reranking with cross-encoder
• Multiple knowledge bases per user
• JWT authentication
• Document preview panel