Reranking

Retrieval often uses a fast first stage to fetch many candidates, then a slower, more accurate reranker to reorder them. Nogueira & Cho (2019) showed that a BERT cross-encoder — which reads the query and a passage together — substantially improves ranking quality over the first-stage retriever alone.

In RAG systems, reranking before passing context to the model improves answer quality by ensuring the most relevant passages are used.