Governance RiskOps Agent for Unity Catalog

WiliamRosa
Databricks Partner

Body:

Every day, data platforms generate thousands of audit events. But here's the problem: security teams are drowning in noise.

Critical risks hide in plain sight. Manual investigations take hours. Compliance gaps surface too late. And there's no intelligent way to prioritize what matters.

I built a solution to fix this.

🚀 Introducing the Governance RiskOps Agent

An automated risk detection system for Databricks Unity Catalog that transforms raw audit logs into actionable security insights.

How it works:

 Continuous Monitoring → Ingests and enriches Unity Catalog audit events in real-time

 Smart Risk Scoring → Multi-dimensional algorithm scores every event from 0-100 using 9 risk factors: • Action type & permission level • Data sensitivity classification • After-hours access patterns • Privilege changes & cross-domain access • Failed attempts & external sources

 Actionable Findings → Not just alerts. Each finding includes: • Exact risk score & severity (CRITICAL/HIGH/MEDIUM/LOW) • Full context (who, what, when, why) • Specific remediation steps

The Architecture:

🏗 Medallion pipeline (Bronze → Silver → Gold) • Bronze: Raw audit event ingestion • Silver: Normalization + dimensional enrichment • Risk Engine: 15+ detection rules with sophisticated scoring • Gold: 4 analytical tables ready for consumption

📊 AI/BI Dashboards with executive metrics (Governance Risk Index, critical findings, risky users)

💬 Genie Space integration for natural language investigation (no SQL required)

Real Impact:

In our demo with 327 realistic events, the system detected: • 86 CRITICAL findings (score 75-100) • 106 HIGH risk events (score 50-74) • 105 MEDIUM risk events (score 25-49)

Investigation time: from hours to minutes.

Production-Ready:

Deploys with Databricks Asset Bundles in a single command Open-source and enterprise-ready Works today with your Unity Catalog audit logs

🎥 Watch the 5-minute demo video to see the full solution in action → [Link to video]

💡This project was built for the DAIS 2026 Community Virtual Contest.

Wiliam Rosa
Data Engineer | Machine Learning Engineer
LinkedIn: linkedin.com/in/wiliamrosa