Scaling critical processes in the energy sector leads to cost optimization and reduced environmental impact. These systems are a complex mix of many software and hardware subsystems, whose finely tuned cooperation is key to successful deployment. In these complex solutions, there is the potential for unpredictable situations to arise, which are often unique and lead to threatening situations or economic opportunities.

The accuracy of implicit solutions for detecting these events decreases significantly with increasing system size. This creates opportunities for the use of machine learning methods. However, the uniqueness of events limits their capture in training data, which is crucial for the deployment of these systems.

With the advent of large language models (LLMs), it has become possible to use artificial intelligence without explicit training for a given problem. These models have a semantic overview that allows them to escape the limiting assumptions of training data and operators.

The goal of this project is to design an LLM-based infrastructure that will clarify unpredictable phenomena and recommend maintenance tasks and interventions for safe management in a given situation. The infrastructure will use not only text but also image and audio recordings to increase contextual awareness. The system will be validated on real industrial equipment in order to increase its safety and eliminate downtime.