I Break Things, AI Fixes Them: Building a Self-Healing CI/CD Pipeline
Talkat Open Source Summit India 2026 · on Tuesday, 2026-06-16· at 15:40· Lotus Ballroom 3 (Level 3)·
summary
Claims a working proof-of-concept called Pipeline Doctor that uses GitHub Actions to trigger a LangGraph orchestration with LangGraph + Gemini for analysis and patch generation, and validates fixes inside a Docker sandbox validation before creating a PR plus auto-generated RCA. Demo intentionally breaks a FastAPI CI; agent fetches logs, iterates fixes, runs tests in the sandbox, and opens a passing-checked PR, cutting human triage from hours or days to minutes. Limits: not for architectural or infra failures.
AI-written summary · about
chapters (13)
- 0:00Self-healing CI/CD intro
- 1:02What is CI/CD
- 3:09Manual CI/CD failure recovery
- 5:21Static workflows vs AI
- 6:03Agentic CI evolution
- 7:50Traditional vs self-healing CI/CD
- 9:56Tech stack for self-healing
- 11:18LangGraph for orchestration
- 13:51Safety in self-healing
- 15:15Self-healing pipeline architecture
- 19:11AI-driven CI/CD safety layers
- 21:14Live demo: break and fix
- 27:37Q&A
related talks
21mAI-driven optimization of CI/CD pipelines
36mAgentic CI/CD: from Pull Request to Production without the Paper Cuts
42mThe Self-Healing Pipeline: How Dagger + LLMs Are Making Traditional CI/CD Obsolete
45mSelf-Healing Pipelines? Resilient Deployments with Hardened Containers
42mProtecting CI/CD from malicious deploys
27mIntegrating Agentic AI Into Internal Developer Platforms To Minimize Errors and Downtime
29mOpen-Source Framework for AI-Powered CI/CD Pipelines
21mBuild your JS Pipeline in Incremental Fashion