Announcing GA of Genie Ontology

Genie Ontology is now Generally Available, offering an agentic layer that continuously learns enterprise context from existing systems such as dashboards, SQL queries, notebooks, and semantic models, while allowing teams to govern critical business definitions. The system captures reusable business context as Ontology Snippets, which update automatically as underlying sources evolve, addressing the limitations of manually maintained ontologies that quickly become outdated. Governed knowledge includes certified KPIs, business domains, and authoritative definitions stored in Unity Catalog, ensuring precision where required. Agents use this shared infrastructure to avoid conflicting interpretations of business rules, improving consistency across deployments.

Agents extract business context from everyday work artifacts, including Databricks dashboards and Power BI semantic models, converting them into reusable knowledge snippets that grow more useful over time. This learned knowledge is combined with explicitly modeled definitions, such as Metric Views and Domains, which are curated and certified by data teams. The system ranks knowledge by authority using OntoRank, an algorithm that weighs factors like source certification, recency, and team expertise to resolve conflicts and deliver trusted context to agents. This reduces redundant exploration and token usage, enabling faster, more accurate reasoning.

Genie Ontology’s effectiveness is demonstrated in a benchmark developed with Turing, which tested agents on a simulated Databricks workspace for a fictional UK energy supplier. Genie, powered by the Ontology, answered 91% of questions correctly—1.6 times the rate of the leading frontier agent—while using less than half the tokens and finishing three to four times faster. The difference stemmed from Genie’s access to pre-learned business context, whereas frontier agents had to rediscover rules on each run, often producing incorrect answers. The benchmark methodology and task walkthroughs are detailed in a research blog, and customers can work with their account teams to replicate the evaluation.

To broaden enterprise context, Genie Search is now Generally Available, enabling agents and users to find and synthesize knowledge from documents, tickets, and wikis alongside governed Databricks assets. The tool respects existing permissions and returns cited, synthesized answers that span multiple sources. Indexed content from these external systems also informs OntoRank, integrating non-lakehouse knowledge into the Ontology’s authority ranking. Administrators configure Genie Search through the admin console, connecting enterprise systems to the Ontology while maintaining governance and security standards.