Urvi Chintukumar Sukharamwala, Daehan Won
The increasing complexity and scale of modern supply chain networks necessitate advanced decision-support systems that are both powerful and interpretable. Traditional optimization techniques, while effective in generating optimal solutions, often produce opaque outputs requiring deep technical expertise to interpret and modify. This research proposed a novel decision-support framework that integrates Large Language Models with mathematical optimization solvers to bridge this gap. An LLM4SCM framework allows supply chain professionals to interact with complex optimization models through natural language, enabling intuitive query processing, dynamic constraint injection, and automated scenario analysis without requiring programming expertise. The methodology includes Named Entity Recognition for extracting decision-relevant information, prompt engineering for aligning natural language queries with solver requirements, and an iterative explanation module for refining model outputs. Through a series of benchmark evaluations and scenario validations, the framework demonstrates its ability to enhance interpretability, scalability, and adaptability in strategic and tactical planning. The results show that LLMs can facilitate transparent decision-making, reduce dependency on technical experts, and accelerate scenario analysis workflows in supply chain management.