Understand this story
Databricks studied how governed knowledge and curated Genie Agents work together to improve the accuracy of Genie One’s answers. In a test of 125 questions for a fictional insurance company, accuracy reached 93%–95% when the two were combined, compared with 83% when either was used alone and 64% when only tables were used. Metric Views, Domains, Subdomains, and Pages primarily improved questions that depended on business definitions, while Genie Agents supplemented information from unstructured documents such as PDFs. Another change test showed that after updating just one Metric View and its Page, all 12 related answers adopted the new definition without requiring any agent modifications. There are currently no facts about subsequent developments.
AI-generated from public reports
Latest development · Report
Databricks Tests Genie One Accuracy, Achieving 93–95% by Combining Governed Context with Curated AgentsDevelopments
1 confirmed developments
Databricks Studies the Impact of Governed Knowledge on Genie One’s Accuracy
Source:Databricks Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 20
Databricks BlogFirst-partySelected
Databricks Tests Genie One Accuracy, Achieving 93–95% by Combining Governed Context with Curated Agents
Databricks tested Genie One on 125 questions about a fictional insurance company. Accuracy reached 93–95% when curated Genie Agents were combined with governed business context from Unity Catalog, compared with 83% when using either one alone and 64% when using tables alone. Metric Views, Domains, Subdomains, and Pages primarily improved questions that depended on business definitions, while agents supplemented information from unstructured documents such as PDFs. In a separate change test, updating just one Metric View and its Page caused all 12 related answers to adopt the new definition, without any changes to the agents.
AI Trends