Writing AI/BI Genie instructions so people can find data

AI/BI Genie is a natural-language, conversational interface in Azure Databricks that lets business users ask questions of curated data and get SQL-generated answers and visualizations. Genie spaces are configured by data owners who attach specific datasets and write instructions to guide the AI's interpretation of schema, terminology, and business logic, improving answer accuracy and discovery. Configuring good instructions is essential for reliable, trustworthy self-service data discovery in a governed environment.

1 · Learn the must-know

• 1A Genie space is created by selecting one or more tables/views (registered in Unity Catalog) that Genie is allowed to query, scoping what data users can discover.
• 2Instructions are written in natural language and can define business terminology, preferred column usage, join logic, filters, and example question/SQL pairs to steer Genie's query generation.
• 3Adding curated sample questions and trusted SQL answers (few-shot examples) within a Genie space significantly improves the accuracy of generated queries for common business questions.
• 4Genie relies on table, column, and comment metadata in Unity Catalog, so well-documented schemas (clear names, descriptions, comments) directly improve Genie's ability to interpret data correctly.
• 5Access to a Genie space and the underlying data is governed by Unity Catalog permissions, so users only see and query data they are already authorized to access.
• 6Genie-generated SQL and results are auditable and can be reviewed/edited by users with appropriate permissions, allowing curators to refine instructions based on observed misinterpretations.

2 · Check your understanding