This project implements a self-improving language model evaluator using synthetic data, inspired by the paper "Self-Taught Evaluators". The model iteratively generates, evaluates, and fine-tunes itself using its own synthetic data, eliminating the need for costly human annotations.