179 questions
No questions match those filters.
What is Anthropic's Contextual Retrieval technique, and...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansA chunk in isolation frequently loses the context that made it useful in the first place — a sentence like “revenue increased 12%” is meaningless without knowing which company, which year, which quarter it refers to. Contextual Retrieval addresses this at index time rather than query time: before embedding a chunk, an LLM reads the full source document and writes a short blurb explaining where this chunk sits and what it’s about, and that blurb gets prepended to the chunk before indexing.
The reported effect stacks with other retrieval improvements rather than replacing them — contextual embeddings alone reduce retrieval failures meaningfully, adding BM25 hybrid search on top reduces them further, and adding reranking on top of both compounds the improvement again.
The economics work because the context-generation cost is paid once per chunk during indexing, not once per query — with prompt caching, generating that context for millions of document tokens is cheap relative to running it live on every retrieval.