Lab: Hollin's embedding decision, measured and switchable

Module 4 · Lab · About 70 minutes · Needs: lessons 1 to 4 of this module

What you are building

Grace wants one page she can sign: which embedding model Hollin Answers uses, how well it finds Hollin's documents, and how to change it later without an outage. Here the lessons' pieces become three files and a command. hollin/embeddings.py holds three pinned local models, a version on every vector, a cache, versioned collections and the hollin-docs alias. evals/retrieval.py scores any ranking on the 32 answerable questions. evals/embedding_decision.py measures four candidates against the live model, applies Lesson 4's gate and writes the decision to evals/results/embedding-decision.json. No language model is called; Ellie's question, "what does E7 mean on my X200" (q19), is the running example.