Lesson · 12 min
Retrieval, not a bigger paste
Fetch the slice, then ask.
RAG (retrieval-augmented generation) means: look up the relevant chunks, put those chunks in the prompt, ask the question, cite the chunks. It is not 'embed the company and hope.' Bad retrieval plus a fluent model is a confident wrong answer with a footnote.
- Chunk on meaning (sections), not on a random 512-token razor.
- Show citations the user can open.
- If nothing retrieved is relevant, say so. Do not answer from the void.