How It Works
Reranking
A second pass after retrieval: a slower, more careful model re-scores the documents the retriever found and puts the most relevant ones at the top before the AI uses them. It trades a little speed for noticeably better answers.
Origin · no single documented coiner
A long-standing technique in search engines and information retrieval; neural rerankers took off after Rodrigo Nogueira and Kyunghyun Cho's 2019 paper using BERT to rerank passages. No single coiner of the word.