Retrieval
Also known as: retrieve
Retrieval is the step of finding the most relevant pieces of information for a query — before the model answers.
In RAG it usually means:
- Embed the query.
- Find the nearest document chunks.
- Hand those to the model as context.
Retrieval quality is the ceiling on answer quality — a model can’t cite what it wasn’t shown.
The underrated skill: great RAG isn't about the LLM — it's about *retrieval*. Better chunks, better embedding model, better ranking = better answers, often more than a bigger LLM.