About Me

I build the systems that make AI engineers trustworthy.

Most people use AI coding assistants. I engineered the operating system mine runs on. In about eleven weeks I designed and shipped a robust AI engineering workspace: 860+ commits, 150+ tooling scripts, 100+ test suites, and a runtime of hooks that stops a bad action before it happens instead of cleaning it up afterward.

What I actually do

I turn judgment into infrastructure. When an AI agent makes a mistake, I don't just correct it. I log it, classify it, and fix the rule that let it through. My workspace keeps a ledger of every caught violation (thousands of entries) and every stated preference, and it feeds them back into its own rules. The system gets measurably better every week, not just more prompts.

I take research and make it production grade. I started from a published methodology for structuring AI context, found where it breaks under real workloads (it assumes a patient human reviewer and free tokens), and built the missing layer: enforcement, execution trails, testing, and token budgets. Four pillars, and every change has to clear all of them.

I test the tests. Every script ships with a test, and the tests ship with mutation checks that prove they fail when the code is broken. A green check that can't go red isn't a check.

I build for security by default. Company code sits behind a walled access layer with an explicit verb set. Private material never leaves the machine in a query. Access controls are treated as answers, not obstacles.

I care about cost. Context windows are budgeted and hook-enforced, repeat work becomes a script, and big outputs get summarized instead of pasted. Efficiency is never bought by skipping verification or tests.

How I work

In one line

I'm a backend engineer who treats AI agents like any other production system: specified, tested, monitored, and held to a standard.