DecisionBox AI Agent ran 233 SQL queries on Databricks in 76 minutes. Here's what it found.

Hey Product Hunt!

I'm Can, co-founder of DecisionBox. I wanted to share a real case study of what our open-source AI agent does when you point it at a data warehouse.

We connected to a Databricks SQL Warehouse with a vacation rental marketplace dataset (11 tables, 500K+ rows). The agent ran a 3-phase discovery:

Phase 1 - Exploration (92 queries): The agent sampled every table, then progressively wrote more complex queries to map the business. When it spotted anomalies, it drilled into root causes autonomously.

Phase 2 - Analysis (74 steps): Structured the findings into 67 insights across 7 business areas, each with severity rating, affected count, confidence score, and supporting evidence.

Phase 3 - Validation (67 steps): Wrote a fresh, independent SQL query for every single insight and re-ran it against the warehouse. 21 confirmed exactly, 12 adjusted with corrected numbers, 1 rejected.

The findings: $5.8M revenue leak from cancellations, 80% host onboarding failure, 43.7% of bookings stuck in pending, 1.84% occupancy vs 65% target.

What makes this different: the agent doesn't generate reports from templates. It reasons about data the way a senior analyst would - forming hypotheses, testing them, following threads across tables, and validating its own conclusions.

Full walkthrough: https://decisionbox.io/blog/databricks-revenue-leakage-discovery

DecisionBox is open source (AGPL v3) and connects to Databricks, Snowflake, BigQuery, Redshift, and PostgreSQL.

GitHub: https://github.com/decisionbox-io/decisionbox-platform