Reranker

A first search stage casts a wide net and returns a rough set of candidates. A reranker then reads each candidate together with the query and scores how well it actually answers it, reordering the list. Because it looks at query and document jointly, it judges relevance more accurately than the first pass can.

Adding a reranker is one of the highest-return steps in a retrieval pipeline. It reliably lifts how often the right passage lands in the top few results, which is where the language model will actually use it.