Staff software engineer, ~14 years. Most recently applied-AI and platform work at Better.com (fintech). I gravitate to the ambiguous, cross-cutting problems nobody owns, and I'd rather build the tool that removes the friction than live with it.
These days I mostly build with AI agents, and I mean that literally: I run a self-hosted multi-agent coding platform (Adjunct) on my own Kubernetes cluster, a seven-agent team of Claude Code and Codex shipping 1,300+ PRs behind human-approval gates. The hard parts (review, guardrails, knowing when to trust the thing and when to pull the plug) are the same judgment calls I've made my whole career.
- mac-scaleset: autoscaling GitHub Actions runners on a pool of your own Macs. An "ARC-for-Mac" built on GitHub's
actions/scalesetand Tart (ephemeral, scale-to-zero, JIT runners). ARC is Linux-only and self-hosted Macs have no autoscaler, so I wrote one. - Adjunct: the platform above. A TypeScript/LangGraph workflow engine, MCP plugins, a policy-gated side-effect broker, and a Next.js dashboard. Private, but I'm happy to walk through it.
- book-of-the-fallen: an NLP and knowledge-graph system over a 3.3M-word fantasy series. BookNLP plus embeddings into Postgres, served to LLMs over MCP, because I wanted to ask it questions no search bar could answer.
Ruby/Rails · TypeScript/Node · Go · Python · React Native · Postgres · Kubernetes · LLM/RAG · MCP
Consulting through Eudemonia, LLC. Open to Staff/Principal roles (applied AI, developer platforms, full-stack), remote.