How I built an AI tool that diagnoses CI/CD pipeline failures in seconds

Source: Dev.to

What it does

When a CI/CD pipeline fails, PipelineIQ automatically:

• Captures the error logs
• Sends them to Claude AI for analysis
• Delivers a Slack alert with the exact root cause and fix steps — within seconds

Example Slack alert

🔴 Pipeline Failure: Stripe API connection timeout blocking payment webhooks

AI Diagnosis: The deployment is failing because the application cannot establish a connection to Stripe's API within the 30‑second timeout limit. This is preventing payment webhook processing.

Recommended Fix: Check STRIPE_SECRET_KEY and STRIPE_PUBLISHABLE_KEY in production environment variables. Test connectivity to api.stripe.com from your deployment environment. Increase API timeout from 30s to 60s.

No log diving. No guessing. Specific, actionable steps.

The stack

• FastAPI – Python backend with async support
• Supabase – PostgreSQL database with Row Level Security
• Anthropic Claude API – AI diagnosis engine
• Slack API – Rich block‑based alerts
• Railway – Production deployment
• GitHub Actions – Integration via one workflow step

How the integration works

Add a single step to any existing GitHub Actions workflow:

- name: Notify PipelineIQ
  if: always()
  run: |
    curl -X POST $PIPELINEIQ_URL/api/v1/pipelines/runs \
      -H "X-PipelineIQ-Key: $PIPELINEIQ_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "repo_full_name": "${{ github.repository }}",
        "branch": "${{ github.ref_name }}",
        "commit_sha": "${{ github.sha }}",
        "commit_message": "${{ github.event.head_commit.message }}",
        "workflow_name": "${{ github.workflow }}",
        "status": "${{ job.status }}",
        "started_at": "${{ github.event.head_commit.timestamp }}"
      }'

Every run—success or failure—is stored. Failures automatically trigger AI diagnosis.

The AI diagnosis engine

A FastAPI background task fires Claude when a failure is stored:

async def run_ai_diagnosis(run: dict, org_id: str, supabase: Client):
    insight = await diagnose_from_run(run)
    if not insight:
        return

    supabase.table("insights").insert({
        "severity": insight.get("severity"),
        "title": insight.get("title"),
        "diagnosis": insight.get("diagnosis"),
        "recommendation": insight.get("recommendation"),
        "estimated_time_save_minutes": insight.get("estimated_time_save_minutes"),
        "confidence": insight.get("confidence"),
    }).execute()

    await send_pipeline_alert(insight, run)

Claude returns structured JSON with severity, diagnosis, recommendation, confidence score, and estimated time saved. The whole process runs in under 5 seconds.

What’s next

• Web dashboard with pipeline health across all repos
• DORA metrics (deployment frequency, change failure rate, recovery time)
• Environment drift detection
• Industry benchmarks — how does your team compare?

Try it free

PipelineIQ is in free beta. I’m looking for engineering teams to try it and give honest feedback on what’s missing.

• Website: https://pipelineiq.dev
• API docs: https://pipelineiq-production-3496.up.railway.app/docs

Happy to answer questions in the comments—especially from DevOps engineers who deal with pipeline failures daily.