Back to catalog / Agent workflows

Answer shopping needs with semantic catalog search

Map natural-language shopping needs to BigQuery attributes and Vector Search results, returning verifiable product candidates.

Goal

Map natural-language shopping needs to BigQuery attributes and Vector Search results, returning verifiable product candidates.

Load product titles/attributes and verify indexed fields and filters. [source-google-answer-shopper-with-semantic-catalog-search]
Retrieve vector candidates, filter by price/category/availability, and show rationale. [source-google-answer-shopper-with-semantic-catalog-search]
Verify SKUs and filters; state when availability is unknown instead of inventing it. [source-google-answer-shopper-with-semantic-catalog-search]

Run fixed queries and verify each product ID exists, hard attributes match, and inventory comes from the catalog result.

If a semantically similar item violates a hard filter, apply structured filters before vector ranking; ask a follow-up when no candidate qualifies.

ADK, BigQuery test catalog, Vector Search 2.0 collection, and Gemini; use test products.

Reads an approved product catalog/vector index and calls embedding or generation APIs; it does not access customer accounts, place orders, or edit the catalog.

Run the source on isolated test data, then check outputs and dependency versions before adapting it.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.