I build AI tools end to end, and I bring the people who will use them along for the ride. Most projects start the same way, with a conversation with someone who actually has the problem, long before a line of it gets written down.
That has meant an accessibility platform for disabled transit riders, scoped with disability organisations rather than guessed at. LLM agents and evaluation systems running in production. ML training for a community of more than a hundred people, most of whom had never trained a model.
Those findings, not my assumptions, drive the design.
What pulls me in is work with something at stake. I am comfortable with ambiguity and thin budgets, and quick to separate what genuinely needs to exist from what merely sounds impressive.