Optimization and machine learning should expand human understanding, not automate human judgment away. The best tools I can build are ones that show people options they couldn't have imagined alone — and then let them choose. That means embracing uncertainty, keeping requirements open, and treating the iterative process of design as something to support, not shortcut.

Real problems don't fit neatly into papers. That's why I work across computational fluid dynamics, molecular force fields, urban planning, and community organizing. The methods transfer. The stakes are always real.