Amazon QuickSight Topics leverage AI to query multiple datasets without pre-defined joins
No more pre-joining datasets: QuickSight's generative AI writes SQL on the fly from semantic context.
Amazon QuickSight's Multi-Dataset Topics offer two modes for combining datasets: defined relationships (pre-wired join graph, inner joins only, DAG constraint) and AI-generated SQL (inferred from semantic metadata at query time, supports outer joins, unions, subqueries, and any relationship graph). The latter mode, enabled by the Chat experience, requires data architects to author a 'Semantic Guidance Stack'—dataset-level instructions, topic-level instructions, field synonyms, and descriptions—that the AI uses to generate context-aware SQL.
The post details eight best practices for configuring this semantic layer, such as providing clear field descriptions, using synonyms for business terms, and giving explicit instructions for handling cross-grain comparisons and role-playing dimensions. It also includes a decision framework for choosing between defined relationships, semantic-only guidance, or hybrid approaches. For data architects and BI engineers, these best practices unlock natural-language exploration across multi-fact schemas without upfront data engineering.
- AI-generated SQL supports inner, left, right, full outer, union, and subquery joins, while defined relationships only allow inner joins.
- Semantic Guidance Stack includes dataset-level custom instructions, topic-level instructions, field synonyms, and field descriptions.
- The approach is best for exploratory analytics and ad-hoc questions; defined relationships suit governed dashboards.
Why It Matters
Enables business users to ask complex cross-dataset questions without data engineering, accelerating self-service analytics.