Reranking

Reranking is a second-stage retrieval step that takes an initial set of candidate results and reorders them with a more precise model to put the most relevant items first.

A fast first-stage retriever fetches many candidates, then a heavier reranking model scores each against the query in detail, improving the quality of the final context.

Reranking is a common way to raise the accuracy of RAG systems, since the documents fed to the model are only as good as their ordering.