Full talk · edited transcript

The 52x Multiplier: How Zepto Mastered AI Agent Evaluation at Scale

Deepak Dhankani, Gireesh Sreedhar K P18 Jun 2026

Deepak Dhankani (Zepto) and Gireesh Sreedhar K P (Databricks)

In quick commerce, delivery windows are measured in minutes, which makes AI reliability existential. At Zepto, scaling to 80,000 daily autonomous support tickets required treating evaluation as core infrastructure rather than "vibe-based" testing. The session has two parts. In the first, Deepak presents the Zepto case study: the support agent stack, evaluation of conversational and multimodal agents, and outcomes. In the second, Gireesh gives a master class on the building blocks of AI evals and the framework behind them: the dual loop built on MLflow 3.0 and DSPy.

Continue reading — it's free

Google or a one-time email code. No password.