How we cut a healthcare partner's LLM spend to $0 per chart
RAG with reranking

Cast a wider net, then re-score it.

Instead of trusting embedding similarity alone, this pulls a wider pool of candidate passages, then a cross-encoder reranker scores each one against the question directly — a slower but more accurate second pass. The top three reranked passages each generate their own answer, so you can compare which passage actually answered the question best.

Your document is indexed on the first question; change the text and it re-indexes.
Run time ~15 MB first run for the reranker, plus the embedding and drafting models
INSIGHTS

Related insights

Try RAG

RAG reranking: when a second pass pays off

Reranking is worth trying when your retrieval is close but not reliable. It is not free accuracy, and the only way to know is to measure it on your own questions.

Contact us

Have a problem worth getting right?

Send us the paragraph you'd normally write to a colleague. We'll get back to you within 24 hours.

hello@tathastha.com