Lightweight, model-agnostic library for evaluating LLMs using curated datasets and customizable scoring.