Lesson 1: Keyword and dense search, measured on Hollin's codes
31 min read·13 Oct 2026
Module 6 · Lesson 1 · About 50 minutes · Needs: Module 4 (bge-small, bi-encoders, recall and MRR) and Module 5 Lesson 1 (approximate search); BM25 in brief is in the AI Engineering course, Module 8 Lesson 3
The situation
On 14 September Hollin recalled X200 jars from batch B-2291. The batch number is printed on the base of the jar, so that is what customers read out and agents type: "B-2291". Hannah Price, who owns the recall, typed it into Module 4's dense retriever. The recall notice came third, behind Calder Fix's pump email and a VAT invoice answer, all scoring 0.52.
Grace wants keyword search added, but asks what it would lose: agents also type "can I use my kettle on holiday in America". This lesson builds keyword search and measures both kinds on Hollin's questions. The Hollin data is simulated.