LLM Evaluation, Synthetic Data & LLM-as-a-Judge

Generation quality evaluation, automated judge models, synthetic-data filtering, faithfulness, and groundedness.

An increasing share of my current research focuses on how to measure quality—not just generate content—including factuality, answerability, groundedness, faithfulness, and instruction quality.

I use LLM-as-a-Judge methods, automated filtering, human validation, and agreement metrics.