Session

Supercharging Genie Performance With Metric Views

Overview

ExperienceIn Person
TrackAnalytics & BI
IndustryEnterprise Technology
TechnologiesGenie
Skill LevelIntermediate
DOWNLOAD SESSION SLIDES

Databricks Genie allows users to talk to their data, but its answers are only as good as the semantic layer beneath it. The Data Platform team shares how we are "supercharging" our internal Genie deployment by feeding it highly structured Metric Views. We will discuss the art of curating data specifically for GenAI consumption, ensuring that natural language queries yield accurate, trustworthy SQL generation. If you want to deploy AI/BI that actually works, you need to see how we structure our data for it.

Session Speakers

Krishna Bhupatiraju

/Sr. Engineering Manager
Databricks

Full Summary

From coin toss to 95 percent accuracy: how Databricks made Genie Spaces work internally

Databricks took Genie Spaces from unreliable novelty to a trusted internal analytics interface that executives now use daily. The breakthrough did not come from a bigger model.

FAQ


A Genie Space is a curated natural language interface built on selected tables, enriched with business context. Users ask questions in English and receive SQL, results, and sometimes visualizations. Answers are grounded only in the included data, not the open web, which keeps outputs constrained and auditable.

Metric views encode formulas, joins, and slicing dimensions, such as a fiscal calendar that starts in February. Genie pulls authoritative logic instead of guessing from table names, which removes an entire class of subtle, recurring errors that prompt engineering cannot reliably fix.

Metric views can still proliferate with conflicting definitions across teams. The metric store enforces a single certified definition per metric using YAML configs and CI checks that fail on duplicates, pushing teams to align before merging changes.

Teams maintain a harness of expected questions, the SQL Genie should write, and the results. A nightly job calls the Genie API, compares outputs to the harness, and alerts the owner when answers drift, typically due to missing comments, stale context, or a changed metric.

Begin small. Clean and comment three core tables, agree on three core metric definitions, and create basic regression tests for expected questions. Only then publish the Genie Space, and keep testing to prevent drift.