RAG Engine SaaS

Production-oriented Retrieval-Augmented Generation (RAG) stack with a FastAPI backend and React frontend.

Overview

This project provides:

• Session-based document ingestion (PDF/TXT/MD/DOCX)
• Chunking + retrieval using embeddings with lexical fallback
• Chat answers with citations
• Optional streaming responses and web-search augmentation
• Multi-provider LLM support (Gemini/OpenAI/Anthropic/Ollama)

Architecture / Stack

• Backend: FastAPI, SQLite, Python 3.11+
• Frontend: React + TypeScript + Vite + Mantine
• Providers: Gemini, OpenAI, Anthropic, Ollama

flowchart LR
  A[Upload Docs] --> B[Chunk + Index]
  B --> C[(SQLite + vectors)]
  D[User Question] --> E[Retriever]
  C --> E
  E --> F[LLM]
  F --> G[Answer + Citations]

Loading

Quickstart

Prerequisites

• Python 3.11+
• Node.js 18+

Install backend deps

cd backend
python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Install frontend deps

cd ../frontend
npm install

Env vars

Create backend env file:

cd backend
cp .env.example .env

Key variables:

• LLM_PROVIDER (gemini | openai | anthropic | ollama)
• GEMINI_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY
• ENABLE_STREAMING (default true)
• ENABLE_WEB_SEARCH + TAVILY_API_KEY (optional)

Run

Backend

cd backend
source .venv/bin/activate
python -m uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Frontend

cd frontend
npm run dev

Frontend default URL: http://localhost:5173 Backend default URL: http://localhost:8000

Test

Backend:

cd backend
pytest

Frontend checks:

cd frontend
npm run build

Deployment

• Backend can be containerized and deployed as ASGI app.
• Frontend builds as static assets with Vite.

# frontend
npm run build

For architecture and operational notes, see docs/.

API snippets

• POST /api/sessions
• POST /api/sessions/{session_id}/files
• POST /api/sessions/{session_id}/chat
• POST /api/sessions/{session_id}/chat/stream
• GET /api/models

Example request (POST /api/sessions/{session_id}/chat):

{
  "message": "Summarize the uploaded document",
  "temperature": 0.2
}

Example response shape:

{
  "answer": "...",
  "citations": [
    {
      "file_name": "example.pdf",
      "score": 0.87,
      "snippet": "..."
    }
  ],
  "used_embeddings": true,
  "model": "gemini-2.5-flash"
}

Troubleshooting

• No model responses: confirm provider API key and LLM_PROVIDER value.
• CORS issues: verify CORS_ALLOW_ORIGINS in backend env.
• Low retrieval quality: inspect chunking settings and uploaded file quality.
• Streaming issues: verify reverse proxy supports SSE.

Changelog

See CHANGELOG.md.

License

MIT — see LICENSE.