Built a RAG system to turn my herbal remedies book into an AI chatbot
I recently launched Herbal Remedies Guide, an AI-powered Q&A system built around my book "The Little Handbook of Natural Remedies."
Built a RAG (Retrieval-Augmented Generation) system that:
- Processes my book content into semantic chunks
- Uses sentence transformers for vector embeddings
- Retrieves relevant context based on user questions
- Generates accurate answers with an LLM (Mistral)
Tech stack:
- Backend: Python + FastAPI with WebSocket support
- Frontend: React + TypeScript + Vite
- ML: Sentence Transformers for embeddings
- Deployment: Vercel
Business model: 3 free queries, then $5 one-time payment for unlimited access
What I learned:
- RAG systems are powerful but tuning the semantic search is tricky
- WebSocket chat feels way better than REST API for this use case
- Lazy-loading ML models saves significant startup time
