Scenarios on Demand: Applying an LLM to Parameterize a Logistics Optimizer at Minerva Foods

This paper presents a framework that integrates Large Language Models (LLMs) and logistics optimization, aiming to improve the export planning process at Minerva Foods. The proposed tool enables scenario generation via chat, allowing users to interact and request new scenarios in natural language. The framework implements a modular architecture that simplifies natural language processing and enhances error tracing and handling. In this phase, we developed a proof-of-concept prototype and evaluated the performance of both gpt-3.5-turbo and gpt-4 models, which demonstrated robust performance in simple tasks as well as in complex scenario analyses derived from real-world cases. In a set of 45 questions, the framework achieved an average error of approximately 3.5\%. With an interface that translates user requests into quantifiable parameters, we aim at making optimizers more accessible and effective, promoting data-driven decision making.