AI-assisted operations
Phoenix AI
An AI assistant I built and operate to investigate pipeline failures and answer infrastructure and common DevOps questions.
- pipeline failure categories
- 6
- unit and property-based tests
- 50+
- available to the engineering team
- Team-wide
01
Problem
A failed pipeline can leave an engineer searching through job logs and platform documentation. Phoenix AI brings that context into the investigation and suggests a practical next step.
02
What it does
Phoenix AI analyzes GitLab CI/CD job logs, classifies pipeline failures into 6 categories and suggests a fix. It can automatically retry jobs after transient infrastructure failures.
It answers infrastructure and common DevOps questions using documented platform context, giving engineers a starting point for investigation.
03
How it is built responsibly
Investigation uses read-only, scoped access. Automatic retries are limited to transient infrastructure failures; engineers remain responsible for investigation and operational decisions.
Features fail safely so a failure in one investigation does not take the assistant down. Changes are checked with 50+ unit and property-based tests and deployed with zero-downtime rollouts.
Phoenix AI is live for the engineering team.
04
Result
Resolved recurring authentication interruptions that had affected the assistant’s availability.