Integrating AutoML and LLMs to streamline the optimisation of production processes. A case study on GAMHE 5.0 pilot line

Optimising production processes is a key challenge in modern industry, with Machine Learning and Artificial Intelligence playing an increasingly important role. Yet, current approaches often require manual intervention and expert knowledge, slowing adoption. To overcome these limitations, this work proposes a methodology that integrates Automated Machine Learning with Large Language Models to carry out process optimisation in a highly automated, end-to-end manner. LLM are employed for code generation, enabling the creation of datasets from existing sources while accounting for contextual information expressed in natural language. AutoML then builds predictive models of the production process, which serve as objective functions in an optimisation algorithm. The methodology was validated through a case study involving four setups, consistently producing optimal solutions aligned with the defined goals. A comparative evaluation of thirteen LLMs for code generation and four AutoML frameworks for modelling was conducted. The best-performing LLMs, DeepSeek DeepSeek-R1, Anthropic Claude Sonnet 4, OpenAI gpt-oss-20b, and OpenAI GPT-5, achieved a 100% functional score. For modelling, TPOT and AutoGluon produced accurate results with coefficients of determination above 0.96 for all the target variables while yielding reliable optimisation solutions.