I model human perception with AI vision models and design the human-evaluation framework for an open-source benchmark measuring how closely those models replicate human visual perception.

The questions behind it are simple to state. How closely do AI vision models replicate what people actually see, and how do you measure perception rather than accuracy. The benchmark I help build aims to make that measurable, so evaluation reflects human visual experience instead of task scores alone.

I also work with Professor Wei Cai at Columbia Business School on simulated research participants, AI models that stand in for human participants in behavioral research, and on NLP research that measures organizational capital and customer service quality from corporate data.