We propose OptiAgent, a multi-agent framework that can output solver-ready mathematical formulations and executable code from natural language descriptions of Operations Research problems. Our architecture prioritizes the mathematical modeling step, where dedicated agents extract structures such as decision variables and constraints, enabling iterative self-correction. We introduce a novel multi-loop validation architecture with four specialized feedback mechanisms targeting distinct failure modes, including misinterpretation, structural defects, mathematical inconsistencies, validation failures, and code errors.
In addition to accuracy, our modular design enhances the optimization problem-solving process by improving transparency, as each agent exposes its reasoning and feedback, making the full modeling process auditable. Our framework achieves state-of-the-art performance on 3 out of 4 benchmarks across LP, MILP, and Nonlinear Programming tasks, while remaining highly competitive on the remaining dataset.
Blogger's Review: OptiAgent's multi-agent design showcases a novel approach to solving Operations Research problems, enhancing interpretability through modularity and transparency. Its potential and challenges in practical applications, especially regarding maintaining performance and accuracy in the face of complex constraints and large-scale data, are particularly noteworthy.