Playground Checkpoint
Design and build a real schema — no auto-grading, just a real attempt.
The Challenge
Six chapters of real vector search, all converging on one real challenge: wire up GreenMart's actual shopping assistant, the way this whole Act argued it should work — matched by meaning, not by shared words.
Open the Playground and work with it directly: run a query with zero shared vocabulary with its correct answer and confirm the real similarity scores rank it first, then add a product of your own to the catalog and write a query that should find it by meaning alone.
What Your Shopping Assistant Needs
- Run the starter query ("something to keep my drink hot") and confirm which product it correctly ranks first — by meaning, sharing no real vocabulary with the winning product's own text.
- Add at least one new product of your own to the Documents file, directly in the Playground (an id and a text field) — then write and run a query that shares no real words with your product's own text, but should still find it by meaning.
- Compare the real similarity scores of your top match and at least one clearly unrelated product — explain in your own words why the gap between them makes sense.
- Note, in your own words, which parts of a real production shopping assistant this checkpoint's Playground can't demonstrate (an actual ANN index at scale, metadata filtering, or the generation half of RAG) — and why each one is honestly out of scope for this specific teaching tool.
Stuck? A Few Hints
- Reread this Act's own early chapters — "cozy jacket for cold weather" scoring 0.6770 against "Insulated Winter Coat" is the exact same shape of result you should expect from your own query.
- Every document just needs an id and a text field — write real, natural sentences, the same way the default catalog does, not just a product name alone.
- The similarity score is shown right next to each result — you don't need to compute or guess it, just read and compare the real numbers the model actually returns.
Ready to Build It?
Opens the Playground, right in your browser — nothing to install.
The starter query should rank the Ceramic Coffee Mug first, with a real similarity score meaningfully higher than the other four products — despite sharing no exact words with its own description. That's the checkpoint working correctly, not a coincidence.
Before You Move On
A query with zero shared words correctly finding its real match, a product you added yourself becoming genuinely findable by meaning, and a real, honest sense of what this Playground can and can't demonstrate at production scale — that's the whole Act, in one real search: traditional databases search for matching data, vector systems search for data that's mathematically close in meaning. The next Act moves from finding the right data to keeping many people looking at the same data in sync, in real time.