PipeMedic

🚑 Inspiration

Debugging failed CI/CD pipelines is a frustrating and time-consuming process.

Developers often have to manually:

• scan logs
• identify the failing job
• correlate errors with recent commits

This slows down development and breaks workflow momentum.

We wanted to solve a simple but high-impact problem: What if pipeline failures could explain themselves?

---

💡 What it does

PipeMedic is an AI-powered GitLab agent that diagnoses failed pipelines by analyzing:

• failing jobs
• job logs
• recent commits and diffs

It provides a structured response including:

• Failure summary
• Likely root cause
• Evidence from logs
• Recommended next step
• Whether rerunning the pipeline will help
• Confidence level

Instead of reading logs, developers get instant, actionable insights.

---

🛠️ How we built it

We built PipeMedic using the GitLab Duo Agent Platform.

• Created a custom agent (agent.yml)
• Designed a flow (flow.yml) to analyze pipeline context
• Integrated GitLab tools:
  • get_pipeline_failing_jobs
  • get_job_logs
  • list_commits
  • get_commit_diff
• Designed a structured prompt for consistent diagnosis output

The system works directly inside GitLab using Duo, making it part of the developer workflow.

---

⚙️ Challenges we ran into

• Understanding the correct schema for agent and flow configuration
• Fixing CI validation errors (placeholders, missing fields, schema issues)
• Handling tool integration incrementally to avoid runtime failures
• Ensuring outputs were structured, useful, and not overly verbose

---

🧠 What we learned

• How to build workflow-driven AI agents (not just chatbots)
• How to integrate AI into real developer workflows
• Importance of structured outputs for usability
• How GitLab Duo Agents can automate real DevOps tasks

---

🚀 Why it matters

PipeMedic reduces time spent debugging pipelines by:

• eliminating manual log scanning
• providing clear root cause analysis
• guiding developers to the fastest fix

It turns pipeline failures into quick decisions instead of long investigations.

---

🔮 Future improvements

• Automatic fix suggestions
• Issue creation for failures
• Historical failure pattern analysis
• Integration with data engineering tools like Airflow and dbt

---

👨‍💻 Who it's for

• Developers
• DevOps engineers
• Data engineers working with CI/CD pipelines

PipeMedic is designed as a practical AI teammate for real-world development workflows.

Built With

• agent
• ai/llm
• ci/cd
• duo
• gitlab
• platform
• yaml