Skip to content

Autonomous Software Delivery

Built an agentic delivery system that turns production issues into competing fixes, tested pull requests, and reviewed releases.

Autonomous Software Delivery workflow: Issues → LangGraph agents → Code + tests → AI evaluation → Reviewed PR. Agent orchestration · Test-backed evaluation · Human review.

From production issue to reviewed PR · Role: Architect & Engineer

Architecture

  • Telemetry intake and issue deduplication feed a self-hosted Gitea agent hub.
  • Two isolated worktrees generate competing implementations; an independent judge scores correctness, scope, style, and tests.
  • A bounded repair pass addresses test failures. Pull requests retain the review and promotion gate.

Results & scale

  • One automatic repair turn per run
  • Two candidate implementations scored across four review dimensions

Technology stack

  • LangGraph
  • Azure AI Foundry
  • Gitea
  • GlitchTip
  • Python

My contribution

Original agent orchestration and delivery tooling, integrated with self-hosted open-source infrastructure.

Related projects

Build with Fern