I take complex processes and make them simple.

The structure around the model

A model on its own isn't much. The useful part is what surrounds it — the tools it can call, the guardrails that keep it in bounds, the way the whole thing is wired together. That structure is what turns a model into something that can actually do the work.

The agent loop

This is where it runs: the agent calls a tool, gets a response, decides what's next, and goes again. Getting that cycle right — the right tools, clean calls, sensible stopping points — is most of what makes an agent dependable instead of unpredictable.

Making it better over time

Things break, and the skill is knowing where. Evals and a human in the loop catch the failures, then the fix goes to wherever the problem actually lives — the prompt, the context, a tool, a guardrail. Not guesswork. Diagnosis, then the right fix.

I started out closer to the engineering side — an industrial engineering degree, then years implementing systems for customers, working directly with the people who had to live with what got built. That’s where the instinct came from: understand the whole process, and the people in it, before touching a solution.

From there into R&D project management — running the delivery side, Agile and Scrum, keeping teams moving and shipping without the process turning into overhead. Then into product, which is where I wanted to be all along: not just delivering what was decided, but deciding what’s worth building.

What keeps me here is taking something tangled and making it simpler, and lately doing that with AI. I like being hands-on: not handing off specs and hoping, but building and seeing if the thing actually holds up.

I’m a people person more than a spec person. I listen, I learn quickly, and I’m comfortable when things are unclear and there’s no obvious right answer.