Reranking re-evaluates and reorders the candidate answers retrieved by a search engine or a RAG system, keeping only the most relevant ones. By filtering informational noise as close as possible to the user's intent, it markedly sharpens the quality of the final answers.
In practice at Gensai
Gensai adds reranking steps to its RAG pipelines when the volume of documents calls for it.