Reranking

First-stage retrieval optimizes recall — getting the right documents into the shortlist. A reranker then reads the query together with each candidate and scores how well it actually answers the query, reordering the top results.

Reranking trades latency for precision, so it applies to the top of the list only. In AACSearch one parameter sends the leading hits through a managed cross-encoder or a small language model, and the search still answers if the model does not.