Building Closed-Loop Evals for a Multimodal Agent at Scale — Soumya Gupta & Jai Chopra, Uber
Uber built a multimodal AI agent to enhance merchant food photos at scale without looking AI-generated
“we're threading the needle here. We need to be able to stay faithful to the original image, preserve the brand of the merchant, and avoid everything looking the same.”
Uber's computer vision team is running a production multimodal agent on Uber Eats to automatically enhance food photos for small merchants who lack time, expertise, and budget for professional photography. The core engineering challenge is threading the needle between AI enhancement and consumer distrust of AI-generated imagery, while maintaining marketplace diversity across 10,000 cities. The talk promises closed-loop eval methodology but the transcript cuts off before that substance is delivered.