Databricks Blog·· 16 hr agoAI score58
Genie One Accuracy: Why Curated Agents Need Governed Context
Genie One Accuracy: Why Curated Agents Need Governed Context
AI summary
A Databricks study of Genie One found that tables alone scored 64% on a 125-question insurance benchmark, while curated agents and a governed layer (certified Metric Views, Domains, Subdomains, and Pages in Unity Catalog) each reached 83%, and combining both reached 93–95%. The governed layer fixed definition questions (100%) and reduced reliance on unrelated Pages and decoy tables, while agents remained the route to team documents; a fictional retailer test showed the governed layer raising accuracy from 54% to 82%.
Source: Databricks Blog · databricks.com