RAG Chatbot Backend

RAG Chatbot Backend

Backend service for the research paper chatbot on Nina Roussille’s website.

Setup

  1. Install dependencies: 
    npm install
  2. Set up environment variables: 
    cp .env.example .env
# Edit .env and add your OPENAI_API_KEY
  3. Ingest PDFs (first time): 
    npm run ingest

    This will:

    • Create a vector store in OpenAI
    • Upload all PDFs from ../files/
    • Print the VECTOR_STORE_ID (add this to .env)
  4. Run the server: 
    npm start
# or for development with auto-reload:
npm run dev

Deployment

  1. Create account at render.com
  2. New → Web Service
  3. Connect your GitHub repo
  4. Set build command: cd rag-backend && npm install
  5. Set start command: cd rag-backend && npm start
  6. Add environment variables:
    • OPENAI_API_KEY
    • VECTOR_STORE_ID
    • PORT (auto-set by Render)

Option 2: Railway

  1. Create account at railway.app
  2. New Project → Deploy from GitHub
  3. Select your repo
  4. Add environment variables
  5. Deploy!

Option 3: Vercel/Netlify Functions

See their serverless function documentation for Express apps.

Updating Papers

When new PDFs are added to /files/:

  1. Run npm run ingest again
  2. It will add new files to the existing vector store

Or use the GitHub Actions workflow (see .github/workflows/ingest-pdfs.yml)

Cost Estimates

  • Storage: ~$0.10/GB/month for vector store
  • Ingestion: ~$0.10/GB one-time per PDF
  • Queries: ~$0.15-0.60 per 1M input tokens, ~$0.60-1.80 per 1M output tokens
  • Estimated monthly: $5-20 for moderate traffic

API Endpoints

  • POST /api/chat - Main chat endpoint
    • Body: { message: string, history?: array }
    • Returns: { answer: string, citations?: array }
  • GET /health - Health check