AI Evals for Cross-Functional Teams — Nachiket Paranjape & Swaroop Chitlur Haridas, DoorDash

DoorDash's evaluation workflow lets operations staff build the interfaces they need while engineering maintains shared APIs. The talk's published description gives examples such as grading restaurant menus and reviewing images: staff can use coding agents to build those task-specific screens on stable endpoints. Human judgments then feed a reference set used to calibrate an automated model judge. This connects the people defining acceptable output with the system that checks it at scale. The platform team supplies common evaluation infrastructure, while the description notes that which group owns a judge prompt varies from team to team.