LLM-Guided Framework for Dynamic Optimization Model Patches

A new research paper introduces an agentic re-optimization framework that uses a large language model (LLM) to act as an operations research (OR) expert, enabling end users to dynamically update optimization models through natural-language interaction. The framework translates user prompts into structured model updates, selects appropriate re-optimization techniques from a toolbox, and solves instances to return implementable solutions. The toolbox leverages primal information such as historical solutions, valid inequalities, solver configurations, and metaheuristics. This approach addresses the challenge of rapidly re-optimizing models in dynamic industrial environments where business rules evolve and unforeseen perturbations occur. The paper is published on arXiv under identifier 2605.18692.