Case studies and notes

The stories behind the résumé lines: what I built, and the calls I made along the way.

i.Case study
I built an AI workflow builder — then deleted it
From an in-app chat builder to a headless MCP plugin inside Claude and ChatGPT; evals, native tool calling, and knowing when the best UI is someone else's.
ii.Case study
Measuring LLM features before trusting them
Deterministic scorers before LLM judges, a simulated user for clarifying questions, grading effects instead of claims, and shadow-testing an LLM's replacement on live traffic.
iii.Case study
Pausing autonomous agents for a human, durably
How to pause an agent before a consequential action and wait days for a human, without losing state.
iv.Case study
ZenGRC Pro: from first commit to a ten-engineer product
Stack choices, a typed GraphQL layer, the Pages-vs-App-Router call, and a redesign with no design team.
v.Method
How I work with agents
The loop, the guardrails, where agents fail, and the numbers.

Letters to the author

Hiring for Staff, Principal or Founding? Write to Kevin.

I built an AI workflow builder — then deleted it