VeriSimpl Boosts LLM Accuracy for Optimization Modeling from Text

Natural language interfaces hold great potential for making complex optimization modeling more accessible, and large language models (LLMs) are increasingly used to convert textual problem descriptions into solver-ready formulations. A critical hurdle, however, is ensuring the generated formulation accurately reflects the user's intent, even if it runs without errors. To tackle this, a new framework called VeriSimpl has been introduced. VeriSimpl employs a novel approach called simplification-based verification. This method involves using the optimization solver itself to create simplified diagnostic queries about a proposed formulation. These queries allow the LLM to more effectively and tractably assess the correctness of its own generated formulation against the original natural language description. Evaluations across various optimization benchmarks show that VeriSimpl consistently improves accuracy over existing methods and provides a high-precision self-verification signal.