Catalog Assistant¶

The Catalog Assistant is a custom-designed assistant for the Community Catalog, available at search.gee-community-catalog.org. Ask it questions in plain language and it returns summaries grounded in the catalog's own pages and dataset metadata — with links back to the source material, so you can verify every answer. You can still use the embedded search within the Community Catalog for basic keyword or text-match lookups.

Open the Catalog Assistant here.

Salient Features¶

• Answers are grounded in the catalog — every summary links back to the source pages and dataset metadata, boosting trust and minimizing inaccuracies.
• Supports multiple languages — ask in your preferred language and the results are summarized in the same language.
• As with all generative assistants, prompt formulation matters. Slight changes in phrasing can often yield better results.

Limitations¶

• Summaries are generated from a limited set of top results, so not every matching dataset may appear in a given answer.
• Summaries may evolve over time as the context window changes, so your answer might vary between searches.

Previous deployment¶

Earlier versions of the assistant ran as a Vertex AI search widget built on RAG (Retrieval Augmented Generation), drawing on two data stores — the Community Catalog pages and a JSON data store — to keep its summaries grounded in reliable sources. That deployment was offered in Beta while we tested and refined it, and its output was a server-side–controlled widget that generated summaries from up to ten search results. Learn more about RAG in this earlier blog post. Notable changes from that phase:

• Added a secondary data store (JSON) to speed up parsing of details like licenses and links.
• Introduced a "Back to the Community Catalog" button for easy navigation back to the main catalog.
• Implemented custom system instructions to improve output quality and relevance.

Ask AI