Give the model your knowledge

RAG fixes the two biggest LLM weaknesses — stale knowledge and hallucination — by retrieving relevant documents and feeding them into the prompt at answer time.

The RAG pipeline

1. Chunk your documents into passages.
2. Embed each chunk into a vector.
3. Store vectors in a vector database.
4. On a question, retrieve the most similar chunks.
5. Augment the prompt with them and generate a grounded answer (with sources).

What makes RAG good

• Sensible chunking and clean source data.
• Good embeddings and retrieval quality.
• Prompts that say "answer only from the context; cite sources."

Takeaway: RAG is how most real-world "chat with your docs" products are built.