Ops-Pilot AI

DASHBOARD
AGENT PAGE
INCIDENTS PAGE
ANALYTICS PAGE
PROJECT PAGE

Inspiration

Modern DevOps teams lose 20–40 minutes per CI/CD failure manually digging through logs, switching between GitLab, Slack, and dashboards just to find a root cause. We wanted to eliminate that entirely.

What it does

OpsPilot AI is an autonomous DevSecOps agent that monitors GitLab pipelines 24/7. The moment a pipeline fails, it automatically:

Fetches and analyses job logs using Gemini 2.5 Flash
Identifies the root cause with a confidence score
Creates a GitLab issue with step-by-step remediation
Opens an auto-fix merge request
Notifies the team on Slack
Triggers the GitLab Duo Agent Platform

All 7 steps complete in 28 seconds — zero human clicks.

How we built it

Google Cloud Agent Builder orchestrates the autonomous agent loop
Google ADK 2.2.0 powers the agent framework with 3 registered tools
Gemini 2.5 Flash handles log analysis, forecasting, and auto-fix generation
FastAPI + Python backend deployed on Render
Next.js frontend deployed on Vercel
GitLab MCP + Duo Agent Platform for pipeline monitoring and issue creation
SQLAlchemy for incident persistence and audit logging
Slack Webhooks for real-time team notifications

Challenges we ran into

Integrating Google ADK with the existing Gemini pipeline without breaking the 7-step agent orchestration
Making the agent truly autonomous — webhook-triggered, no manual intervention required
Getting Gemini to reliably return structured JSON under all failure scenarios
Keeping the full agent run under 30 seconds including GitLab API calls

Accomplishments

Full autonomous 7-step agent pipeline working end-to-end in production
28-second mean time from pipeline failure to GitLab issue + MR open
Predictive risk forecasting using historical incident patterns
Google ADK 2.2.0 integrated with live tool registration

What we learned

Building a truly autonomous agent is harder than building an AI feature. The orchestration, error handling, fallbacks, and timing all have to work perfectly together in production — not just in demos.