---

My path has taken me through backend engineering, machine learning, and now systems where language models are part of the stack.

That path shaped how I think.

I don't approach these problems from the demo outward. I approach them from the failure modes inward: what breaks, what scales, what can be trusted, what needs to be measured, and what happens when real users start depending on it.

My default question is simple:

What is the simplest thing that actually works?

That usually means resisting complexity before adding it. Before reaching for a larger model, a multi-step workflow, or a network of agents, I want to understand the problem underneath: the user need, the edge cases, the cost of being wrong, and the point where automation should stop.

I'm most interested in the space between engineering judgment and emerging AI capability — turning frontier AI into real-world systems that are grounded enough to trust, observable enough to improve, and controlled enough to use in production.