Project Escher
Fellows will author challenging visual and compositional reasoning questions (with verified answers) over real-world charts to support multimodal vision-language model development. The goal is to benchmark model performance in chart and diagram question answering.