Build a docs Q&A (RAG) app

Retrieval-augmented generation (RAG) lets a model answer from your PDFs, notes or wiki. You split documents into chunks, embed them into a vector store, retrieve the most relevant chunks per question, and pass them to the model with the question.

Why this stack

• LlamaIndex and LangChain have ready-made loaders, chunkers and retrievers.
• Chroma runs locally for free; Supabase gives you Postgres with pgvector when you need a hosted database.
• Long-context models like Gemini Flash tolerate generous retrieved context cheaply.

Step-by-step

1. Load 10 representative documents and write 5 test questions with known answers.
2. Index them with LlamaIndex + local Chroma and check the answers.
3. Tune chunk size and top-k until the test questions pass, then build a UI.
4. Show source citations with every answer.

Common pitfalls

• Bad PDF text extraction is the most common failure. Inspect the extracted text first.
• Re-index when documents change.
• Don't expose private documents through a public endpoint without auth.