I build AI systems that help people get better at hard jobs.

Client work

• Search expert profiles from a problem description

  I built a search system across about 200,000 profiles using semantic search, exact terms, domain signals, reranking, and diversification. A trace showed why each result ranked. Read the technical case.

• Let domain experts change pipeline rules

  Domain experts changed complicated rules and ran cases through the pipeline without writing code. When one failed, we separated a code bug from a missing or mistaken rule. That distinction led to Sketch-CE.

• Draft performance reviews and growth plans

  One system combined 360-degree feedback, a skills rubric, and past reviews into a draft performance review and growth plan. Another compared project briefs with availability, prior assignments, expertise, and performance evidence to suggest a team. See the client systems.

Open-source projects

Some projects are open source because I wanted to make them and can share them.

• Repo-Native Alignment

  A local code graph that lets agents ask what a change touches, find code by meaning, and connect a commit to the business reason behind it.

• Sketch-CE

  A paper and code experiment that keeps an agent's code fix separate from a change to the business rule. People approve rule changes; the agent can repair an implementation error without quietly changing the rule.

• Unified Hi-Fi Control

  A Rust bridge connecting four audio systems to a physical knob, web UI, Apple clients, and MCP. I use it every day. It is also fun.

Writing

• Agent Work That Changes Behavior

  What has to surround an agent before its next run can improve.

• AI: Token or GPT?

  Why the words around AI often travel faster than the work they describe.

• When AI Changes the Work, the Complements Change Too

  What a company still needs after a model makes one task cheaper.