Agent & LLM orchestration engineer. I build agent systems in production — planner-led multi-agent orchestrators, MCP tool servers, and the evaluation pipelines that keep them honest — and I run my entire computing life declaratively.

This is my working account — forks, prototypes, protocol walkthroughs, and learning repos. Less curated, more honest.

My complete personal infrastructure as a NixOS monorepo: five hosts (two servers, a Framework laptop, two Android phones via nix-on-droid), flake-parts modules auto-discovered at eval time, sops-nix secrets, ~68 agent skill definitions, and lint/eval/schema validation gates. It's my whole toolbox, publicly available. (Hosted on my other account, magus-john-bee.)

Agent orchestration, RAG, and LLM evaluation at S&P Global Market Intelligence (2023–present). Previously: crypto backends, ERP/platform engineering, teaching (NYC public schools, Peace Corps Indonesia).

Currently exploring: WASI components as sandboxed, hot-swappable agent units.

Interested in: evaluation of AI agents · research software engineering · knowledge graphs · adaptive tutoring · applied ontology