From 6387a84558e267497eb053f9a3a3c1bc1236c870 Mon Sep 17 00:00:00 2001 From: kjannette Date: Thu, 30 Jul 2026 05:51:23 +0000 Subject: [PATCH] Update README.md --- README.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/README.md b/README.md index a6e1e04..a4d79e2 100644 --- a/README.md +++ b/README.md @@ -71,6 +71,14 @@ Archetypical cosine scores range from -1 to 1. 0: Vectors are at a 90-degree angle (orthogonal/unrelated). -1: Vectors point in opposite directions. +## The role of rerank models + +Rerankers are high-precision, fine second (or third) stage filters. Rerankers run "from scratch" on query-source/answer-source pairs and are computationally expensive. + +A reranker assigns relevance scoring by evaluating query-source/answer-source pair similarity within a multi-dimensional vector space. They should be RAG-pipeline-downstream of embedding models, which are "coarse" - and occasionally yield false positives on semantically irrelevant chunks. + +Reranker are justifiable and high-value where 1. accuracy is at an absolute premium because of the magnitude of downside risk (ex. uses cases: litigation, health care and medical research science) 2. token-economization heuristics exist at earlier pipeline stages, and 3. system and network latency reduction is optimized. + ## Design Two-package monorepo: