I've built production agentic systems for Chainguard Inc. that automatically diagnose and fix build failures across tens of thousands of open source packages. These systems use multi-agent orchestration, specialized Claude Code skills for each build tool and error class, and iterative fix-validate loops that run entirely within GitHub Actions pipelines. The result: packages that previously required manual triage are fixed autonomously, at scale, without human intervention. I bring that same approach — decomposing large problems into focused agent tasks, building reliable iteration loops, and integrating agentic workflows into existing CI/CD infrastructure — to client engagements.
Areas of expertise
Agentic workflow design
Design multi-step agent workflows that decompose complex engineering tasks into focused, reliable sub-tasks — reducing the scope of each agent action and improving overall correctness.
Claude Code skill engineering
Build specialized Claude Code skills and sub-agents for domain-specific tasks, with schema-driven inputs, explicit guidance, and hard-stop detection for unsolvable cases.
CI/CD integration
Integrate agentic workflows into GitHub Actions or other CI/CD pipelines so AI operates autonomously on each PR, build failure, or scheduled trigger — no human in the loop required.
Iterative fix-validate loops
Build agent systems that test their own output, validate results against the actual system under test, and iterate with refined strategies — not just one-shot generation.
AI-powered diagnosis and triage
Apply LLMs to analyze build logs, test failures, or error reports and classify them by root cause — enabling automated routing, prioritization, or autonomous remediation.
Training and adoption support
Help your team get up to speed with Claude Code, prompt engineering, and agentic system design — from first experiments to production-grade workflows.
Not seeing what you're looking for?