Researchers have developed a pipeline that uses large language models (LLMs) to generate graphs for optimizing complex problems. The pipeline is problem-agnostic and can be applied to any problem in the MiniZinc format. It prompts an LLM with guidelines to produce a graph generator, which maps decision variables and constraint relationships to nodes and edges. This enables the structure-based local improvement framework to select variables efficiently. Experiments on 20 problems showed that algorithm selection achieved a 39.6% win rate against a baseline, more than doubling the best single configuration. --- Why it matters: This matters because it shows how LLMs can be used to automate and improve optimization methods for complex problems. The ability to generate problem-agnostic graphs could lead to significant improvements in efficiency and effectiveness of constraint optimization algorithms. Source: https://arxiv.org/abs/2608.13333