Reranking
A second retrieval step in which a more precise model reorders an initial set of search results by relevance to the query. It improves the quality of the few passages that are actually passed to a language model.
Why it matters
Initial retrieval is designed to be fast across large collections, so its ranking is approximate. Reranking spends more effort on a short list, which usually improves answer quality in retrieval-augmented generation without changing the underlying index.
It adds latency and cost per query, so its benefit should be measured against the organisation’s own questions and documents.
In practice
For example, a pharmaceutical regulatory team’s assistant might retrieve fifty candidate passages from submission documents, rerank them, and pass only the top five to the model, improving the precision of cited answers.
Where Rodan fits
Rodan tests retrieval pipelines end to end, including reranking, as part of AI and Decision Systems delivery.