I started in software engineering. What kept pulling me toward design wasn't a change of interest. It was the same problem appearing everywhere: systems that worked fine technically but failed the people using them. Gaps between what a product could do and what users could actually understand it was doing.

That gap is still what I work on. The engineering background shapes how I collaborate: I know what "backend constraint" actually means during a sprint review, I've reviewed PRs, and I've sat through enough implementation sessions to know exactly where design decisions get lost in translation. When I say something isn't feasible to ship this sprint, I know why.

I've worked on Keycloak-based authentication systems, which taught me that security requirements and usability requirements aren't opposites. They're a constraint that takes real craft to resolve without sacrificing either. I've designed onboarding flows for underbanked demographics in Brazil, where shared devices, spotty connectivity, and limited digital literacy aren't edge cases. They're the baseline.

Accessibility is where I find the clearest design problems. When you design for someone using a screen reader, or with a motor impairment, or on a 2G connection, every ambiguous decision has to become explicit. The constraints force the work to be cleaner. That turns out to be better for everyone, not as a principle, but as a practical outcome you can see in the finished product.

In 2026 I open-sourced five AI skills for Claude: UX auditing, design critique, structured discovery, strategic design, and frontend specification. I built them because I was frustrated with AI feedback that sounded authoritative but wasn't grounded in anything. Each skill labels what's data, what's estimated, and what's an inference. They're on GitHub.