Abby Feng

Contributing Writer, Explore Agentic

About Abby

Abby writes the evaluation-program side of shipping LLM systems. The long-form LLM evaluation guide is the anchor: what to measure on a generation task versus a retrieval task versus a full agent trajectory, how to build an LLM-as-a-judge that agrees with human labels often enough to be worth acting on, how RAG evaluation differs from end-to-end agent evaluation, and how to wire the whole thing into CI so a regression fails a pipeline rather than surfacing in a support ticket weeks later.

The recurring argument is that the hard part is rarely the metric — it is the program around it: who owns the eval set, how it grows, what happens when a score moves, and how an engineer defends a quality number in a review meeting. The healthcare review byline extends that to voice agents calling insurance payers, where the escalation threshold is the real quality control. Expect rubrics and operating mechanics, not leaderboards.