Wonder Engineer Kartik Arora Discusses Agent Evaluation and Prompt Management
In a video published on September 28, 2026, Kartik Arora, an AI Engineer at Wonder, outlines how his team uses LangSmith for tracing, LLM-as-judge scoring, and prompt management to automate feedback loops in their generative AI application.
ImportanceLocalEvidenceE2 unreplicatedWrite-upQuick
Kartik Arora, an AI Engineer at Wonder, explains in a video published on September 28, 2026, how his team builds a generative AI application that plans and delivers all 21 meals for a user's week. He details the transition from manual log debugging to using LangSmith traces, evaluators, and LLM-as-judge scoring to assess how their agent responds to different users.
Arora notes that implementing prompt management transformed a feedback loop that previously took minutes to hours into an automated process. With a small team of four engineers and two product managers, they now assign bug tickets to a coding agent that reads LangSmith data, locates issues in the codebase, and fixes them end-to-end.
The video chapters also cover enabling non-engineers to own prompts and observing the development of the AI infrastructure ecosystem. This resource offers practical insights for readers interested in agent evaluation mechanisms and operational workflows in production environments.