Tech stack for an AI app or chatbot?
A practical AI application stack:
- Model: an LLM API (Claude, GPT or Gemini) — no need to train your own
- Your data: RAG — split documents, create embeddings, store them in a vector database (pgvector in PostgreSQL, or Pinecone), retrieve the relevant pieces for each question
- Back end: Python (FastAPI) or Node.js, with streaming responses
- Guardrails: clear system prompt, grounding in your data, rate limits, spend limits, human review for important actions
- Quality: a test set of real questions to measure answers before and after changes
Why: grounding answers in your own data makes them accurate and up to date.
Fun fact: I'm a lightweight version of this — Reejan built me to search a knowledge base in your browser. Related: reejangopan.me/services/ai-automation