Databricks · One-Pager · AI / BI

AI/BI Genie & Dashboards

Ask questions of your governed data in plain English — Genie writes the SQL, runs it on Unity Catalog tables, and answers with results and charts; AI/BI Dashboards turn the same data into shareable, AI-assisted reports.

GA NL → SQL on UC data Tagging in Public Preview Verified June 2026 docs

In one line

A Genie space is a domain-specific natural-language chat interface where users ask questions of their data and get back generated SQL, result tables, and visualizations — running on a SQL warehouse over Unity Catalog datasets.

Why it matters

It lets business users self-serve answers without writing SQL or filing a request to the analytics team, while staying inside Unity Catalog governance — Genie can only read tables the asker has SELECT on. AI/BI Dashboards then package the curated, recurring views; the two share the same governed data and semantics.

Core concepts

Genie space

A curated chat surface scoped to up to 30 Unity Catalog tables/views and a SQL warehouse (serverless recommended); data analysts tune it for one business domain.

NL → SQL generation

Genie turns each question into SQL, runs it on the warehouse, and returns rows plus a chart — and shows the generated SQL so it can be inspected.

UC metadata grounding

Genie reads Unity Catalog column names and table/column descriptions to understand the data; quality comments are critical for accuracy.

Instructions & examples

Steer it with three layers — SQL expressions (metrics/filters), example SQL queries, then plain-text instructions as a last resort. Up to 100 per space.

Verified answers

Trusted assets — parameterized example queries and UC SQL functions — return a deterministic, pre-validated answer when a question matches.

AI/BI Dashboards

AI-assisted reports built on datasets (tables, views, or queries) with charts and global/page/widget filters; you can publish a Genie space alongside them.

How it works

Click a stage

1 · Question 2 · Genie + context 3 · Generated SQL 4 · Result / visual

NL question — a user asks in plain English, e.g. "What were total sales last month by region?".

Minimal example

Genie space · steer it, then ask# Example SQL query you register to teach Genie a metric:
# title — "Monthly revenue by region"
SELECT region, date_trunc('month', order_ts) AS mth,
       sum(amount) AS revenue
FROM sales.gold.orders
GROUP BY region, mth;

# User then asks in chat:  "total sales last month by region"
# Genie matches the example, generates aligned SQL, runs it,
# and returns rows + a bar chart — showing the SQL it ran.

Gotchas & limits

Garbage in, garbage out: accuracy depends on good Unity Catalog metadata — clear table/column names and descriptions are critical.
Generation is probabilistic, not deterministic — only verified answers (parameterized queries / UC SQL functions) guarantee the same trusted SQL each time.
Hard limits: up to 30 tables/views, 100 instructions, and 200 knowledge-store snippets per space; throughput is capped per workspace.
Governance, not a bypass: Genie respects UC permissions — a user can only query data they already have SELECT on.

Interview soundbites

Say this out loud

"Genie is natural-language-to-SQL over Unity Catalog data — it generates SQL, runs it on a SQL warehouse, and shows you the SQL so you can trust the answer."
"You steer it with curated context: SQL expressions for metrics, example queries, and text instructions — and you lock in known answers with verified/trusted assets for determinism."
"AI/BI Dashboards and Genie are two faces of the same governed data: dashboards for curated recurring reporting, Genie for ad-hoc questions — and you can publish a Genie space right alongside a dashboard."

References

Verified against docs.databricks.com · June 2026