BUP59 - FitLLM

In modern mobility and production, companies face highly complex planning and decision-making tasks. With constantly changing constraints, there is a need for heuristic algorithms that are both adaptable and powerful and can be used specifically to solve specific optimization problems. Until now, powerful heuristics have had to be developed by experts at great expense – a time-consuming and resource-intensive process.

FitLLM is the first to use small, locally executable language models (Large Language Models, LLMs) to automatically generate customized optimization heuristics and adapt them flexibly to changing mobility and production scenarios. LLMs are AI models based on natural language processing.

Aim

The goal of FitLLM is to develop an optimization framework based on generative artificial intelligence. Using small, locally deployable LLMs, the system automatically generates customized heuristics for solving complex optimization problems in the fields of production and mobility. Context-based inputs and seamless integration into established processes enable both initial solution generation and iterative refinement. The result is a resource-efficient, data-secure, and flexible optimization framework that can be directly transferred to industrial applications.

Approach

• Selection and modeling of realistic optimization problems from production and mobility, and creation of a dataset of representative problem instances
• Generation of high-quality heuristics via context-based prompt strategies with small, locally executable LLMs
• Integration of the generated heuristics into established optimization methods.
• Systematic evaluation of performance based on industry-relevant application scenarios.

Benefit

Only through the automatic generation of tailor-made heuristics with small, locally deployable LLMs is it possible to solve complex planning and optimisation tasks – such as dynamic fleet management or flexible production planning – efficiently, securely and adaptably.

For the end user, this means that processes become faster, resources are used more efficiently and systems respond more reliably to changes.