RAG Chatbot Setup Guide

RAG Chatbot Setup Guide

This guide will help you set up a RAG (Retrieval-Augmented Generation) chatbot that allows visitors to ask questions about Nina’s research papers.

📋 Overview

The chatbot consists of:

  1. Backend service (rag-backend/) - Handles AI queries
  2. Frontend widget (_includes/chatbot.html) - Chat interface on website
  3. GitHub Actions (optional) - Auto-updates when PDFs are added

🚀 Quick Start

1. Set Up Backend Locally

cd rag-backend
npm install
cp .env.example .env
# Edit .env and add your OPENAI_API_KEY
npm run ingest  # Creates vector store and uploads PDFs
# Copy the VECTOR_STORE_ID to .env
npm start  # Test locally

2. Deploy Backend

Choose one:

  • Render.com (easiest, free tier available)
  • Railway.app ($5/month, more reliable)
  • Vercel/Netlify Functions (serverless)

See rag-backend/README.md for detailed deployment steps.

3. Update Chatbot Widget

Edit _includes/chatbot.html line ~180:

const API_URL = "https://your-backend-url.com/api/chat";

4. Test

  1. Push changes to GitHub
  2. Visit your site
  3. Click the “Ask” button (bottom right)
  4. Try asking: “What is Nina’s research about?”

💰 Costs

OpenAI API:

  • Vector Store: ~$0.10/GB/month storage
  • Chat queries: ~$0.15-0.60 per 1M input tokens
  • Estimated: $5-20/month for moderate traffic

Hosting:

  • Render: Free tier (spins down after inactivity)
  • Railway: $5/month
  • Vercel/Netlify: Free tier available

Total estimated: $5-25/month

🔧 GitHub Actions (Optional)

To automatically ingest new PDFs when added:

  1. Go to repo Settings → Secrets → Actions
  2. Add secrets:
    • OPENAI_API_KEY
    • VECTOR_STORE_ID
  3. Push a PDF to /files/ - it will auto-ingest!

📝 Files Created

  • rag-backend/ - Backend service
  • _includes/chatbot.html - Frontend widget
  • .github/workflows/ingest-pdfs.yml - Auto-ingestion workflow
  • RAG_SETUP.md - This file

🐛 Troubleshooting

Chatbot not appearing?

  • Check browser console for errors
  • Verify API_URL is correct in chatbot.html
  • Ensure backend is deployed and running

Backend errors?

  • Check environment variables are set
  • Verify vector store ID is correct
  • Check OpenAI API key is valid

PDFs not ingesting?

  • Ensure PDFs are text-based (not scanned images)
  • Check OpenAI API key has sufficient credits
  • Review ingestion logs

📚 Next Steps

  1. Customize chatbot styling in chatbot.html
  2. Adjust system prompt in server.js for different tone
  3. Add rate limiting (already included)
  4. Monitor costs at platform.openai.com/usage

🆘 Need Help?

Check:

  • rag-backend/README.md - Backend documentation
  • rag-backend/ENV_SETUP.md - Environment setup
  • OpenAI docs: https://platform.openai.com/docs