Large Language Models in Business Practice | Offenburg University of Applied Sciences

A man stands in front of a seated audience in a lecture hall and explains a presentation on the screen behind him

In a brief survey of the approximately 40 attendees and 30 online participants, Prof. Dr. Janis Keuper, director of the Institute for Machine Learning and Analytics at Hochschule Offenburg, made it clear at the outset how the use of AI—particularly large language models (LLMs) and agent systems—has evolved in small and medium-sized enterprises in recent years. Although these technologies have been available for some time, a rapid increase in their use has only been observed over the past year and a half. Janis Keuper explained that the reasons for this include major technical and legal challenges, high costs coupled with benefits that are difficult to evaluate, as well as concerns regarding dependencies and security.

Drawing on preliminary findings from the “LLM-Praxis” research project at Offenburg University of Applied Sciences—funded by the Federal Ministry of Research, Technology, and Aerospace (BMFTR) at Hochschule Offenburg and real-world use cases from regional companies, the six speakers from research and industry then presented potential solutions to these issues. First, Pius Horn, a doctoral candidate and member of the LLM-Praxis team, explained document parsing: the first step toward integrating local information into language models or agent systems. Bianca Lamm, a data scientist at Markant Services International GmbH, then presented “Multimodal Structured Output Prediction,” which involves multitasking with language and vision models. After a short break, Arnt Bieberstein, a doctoral candidate and staff member at LLM-Praxis, first explained how to build chat agents correctly and, finally, how to control tone using ACT. Louis Emir, a master’s student at BadenIT GmbH, discussed the practical application of Recursive Language Models for very large context windows. Tobias Lachmann, Head of Development at Schrempp EDV GmbH, joined Schrempp Managing Director Stefan Basler to present “ERP in Dialogue,” demonstrating how CRM business processes can be controlled using AI and language. And Janis Keuper addressed the topic of LLM security: “How Vulnerable Are LLMs to Manipulation?”

The company representatives in attendance used the concluding “Meet & Greet” to network and gather additional practical tips from the speakers.

The latest results of the university project can also be found at https://www.llm-praxis.de/.

Future.Space.Black Forest

The RegioWIN flagship project Zukunft.Raum.Schwarzwald aims to increase the innovative capacity of small and medium-sized enterprises by promoting structured, needs-based knowledge and technology transfer, and to integrate rural and suburban areas—through coworkingand innovation hubs.