Reranking

Also known as: Reranker

Reranking rescores an initial candidate set of retrieved passages with a stronger (often cross-encoder) model so the LLM sees the best evidence first.

A cheap retriever proposes top-50; a reranker keeps top-5. This two-stage pattern is a high-leverage RAG upgrade when recall is decent but precision at $k$ is weak.

← Back to AI Glossary