AI Evals for Cross-Functional Teams — Nachiket Paranjape & Swaroop Chitlur Haridas, DoorDash
DoorDash evolved AI evals from an engineering task into a cross-functional platform for non-engineers.
“cost is a number one concern these days”
DoorDash's GenAI platform team describes building a horizontal infrastructure layer—LLM gateway, agent gateway, and open-weights model hosting—that helps product teams balance accuracy, latency, and cost. Their core lesson is that evals must be UI-first to empower non-engineers like PMs, ops staff, and labeling partners who are the true domain experts. Different product teams required distinct eval paradigms: session-level quality judgments, scaled human labeling, and trajectory-based evals for multi-agent systems.