AI glossary
Reranking
Reranking takes the first results of a fast search and reorders them with a slower, more careful model so the best passages come first.
A first search returns, say, fifty candidates cheaply. A reranker then reads the query and each candidate together and scores how well that passage answers that question.
Because it reads them in pairs, it is far more precise than comparing two separate embeddings, and far too slow to run over a whole collection. That is why it only sees the shortlist.
Reranking cannot rescue a passage the first search never returned. If the right chunk is not in the shortlist, widen the first search or fix the chunking first.
Learn it hands-on
These free lessons teach it with commands to run in a lab and checked tasks. Read them whole with no account.