Put neural search on your own catalogue

The browser lab has no project backend, API key, database, or training step. A fork can become a small semantic-search demo for your own catalogue in about five minutes.

1. Fork the repository.
2. Replace the 20 sample products in docs/catalog.json with your own public, non-sensitive catalogue data. Keep the same five fields. The exact contract is published as docs/catalog.schema.json, and the lab now reports a readable error instead of silently accepting malformed data.
3. In the fork, enable GitHub Pages with Deploy from a branch, branch main, folder /docs.
4. Open /lab.html on your Pages site and try a query.

[
  {
    "id": "your-product-id",
    "title": "Product name",
    "category": "Category",
    "description": "A concise searchable description.",
    "tags": ["useful", "search", "terms"]
  }
]

The visitor’s browser downloads the pinned public MiniLM model and embeds the catalogue locally. Large catalogues need a different delivery/indexing design; this remix path is intended for small demonstrations and prototypes.

If you publish a remix, open a demo-feedback issue with the link. Independent remixes are useful evidence and help shape a more general catalogue adapter.