Insight-on-demand data agent

Turns plain-English business questions into SQL, runs them against the warehouse, and returns narrated insights and charts for leadership.

Analytics · Stack: Claude, BigQuery, Comet. Outcome: Self-serve analytics for the CEO's office — zero wait

Overview

A text-to-insight agent that answers plain-English business questions against the warehouse with narrated results and charts.

The problem

Leadership questions queue behind the data team, so simple answers take days and the CEO's office waits on a human for every number.

How it works

• Takes a plain-English business question.
• Translates it into SQL against the warehouse.
• Runs the query and validates the result.
• Returns a narrated insight with supporting charts.

What it does

• Natural-language to SQL
• Governed warehouse access
• Narrated insights, not raw tables
• Auto-generated charts

Architecture

Claude for query generation and narration over BigQuery, with Comet for observability on the runs.

Results

• Self-serve analytics for the CEO's office with zero wait
• The data team freed from ad-hoc pulls
• Faster leadership decisions