Lab: Keyword search over the help centre

Module 5 · Lab · About 60 minutes · Needs: lessons 1, 2 and 5 (cleaning and tokens, BM25, error analysis by hand)

What you are building

Priya's request from lesson 2 still stands: when a ticket arrives, attach the help article that answers it. In this lab you build that as one function, search(query), over the 12 help articles. It cleans and tokenises text with lesson 1's functions, scores articles with lesson 2's BM25, and returns the top 3. Then, as lesson 5 taught, you measure it on real tickets, read every mistake by hand and count what one fix fixed and broke. Last, you save the index to a file with a fingerprint, so a stale index cannot be loaded by mistake. No model is called and no key is needed. The running example is a real shared ticket: "Refund still not received after two weeks!!"