Retrieval-augmented generation
Learn why applications retrieve documents before generation.
AI Engineering tool
RAG Retrieval Exercise
Explore retrieval precision, recall, and the context-token cost of adding more chunks.
Runs in your browser with editable illustrative assumptions. It does not call a model, send entered content to a server, or produce a current market quote.
Interactive calculator
RAG retrieval exercise
tokens
Retrieval quality is a balance: include enough relevant evidence without filling the context with low-value material.
Retrieval precision75.0%
Retrieval recall60.0%
Approximate retrieved context2,800 tokens
This exercise uses your labels, not a live vector store. Validate both relevance and grounded answer quality before increasing retrieval depth.
Starting values are illustrative defaults you can edit — not live ComputeTape Market benchmark prices. Replace them with a real quote.
How to use it
Make the assumption visible
Label which retrieved chunks are relevant, then compare precision and recall with the token volume added to the model context.
- Change one assumption at a time when comparing designs.
- Record the prompt, workload, or permission version beside the result.
- Treat the output as a planning aid and validate it against representative work before making a production decision.