This project implements a multi-agent debate system where multiple large language model instances argue and critique each other’s responses. The framework addresses the degeneration-of-thought problem inherent in solo self-reflection by introducing adversarial interaction between agents. Each agent presents arguments, challenges the other’s position, and iteratively refines responses through the debate process, with the goal of converging on more accurate and well-reasoned answers.