InnoLOG: 8th Thuringian AI Forum October 6, 2026 | COMCENTER Brühl | Erfurt

Take the opportunity to discuss exciting questions and share your own experiences. The Cross-Industry Dialogue raises questions, provides answers, motivates, inspires and connects.

In the moderated group discussions of the Cross-Industry Dialogue, developers, decision-makers, users, and interested parties from various domains come together to discuss current questions regarding the use of AI and the associated challenges. How will Thuringia become an AI location? How can integration into existing processes succeed so that AI helps and does not hinder? What risks does the use of AI entail, and how can they be addressed? How does AI support the mobility of tomorrow?

The use of AI models - especially language models from large providers - is not possible in every application context. However, operating large models in one's own or trusted infrastructures requires high resources and may also not meet the requirements. Small yet powerful AI models for use on hardware with limited resources, e.g., on edge devices, appear promising. But which models are the right ones for the application context and how can they be used effectively? The workshop provides insights into typical methods for compressing large AI (language) models. Together, we will take a look at how we can use these models, whether they are suitable for our local conditions, and how we can test models for our use cases.

The topic should be control over one's own data when using AI in industry: Edge processing strengthens this control by processing data locally, close to the point of origin – not in the cloud. Privacy by Design integrates data protection into hardware and software from the outset, so that only generated information leaves the device. Challenges include energy consumption, limited storage, and real-time requirements. The workshop shows solutions such as data reduction, compression, and embedded AI to reduce the data stream on sensors and edge devices, thus conserving resources.

LLM-based chatbots have become the de facto standard for text-based first-level support on websites in recent years. Typically, a Large Language Model (LLM) accesses one or more knowledge sources in the form of vector databases or knowledge/context graphs. However, since such knowledge databases change over time and thus influence the outputs of chatbots, continuous monitoring of these outputs has become all the more important. In the workshop, Prof. Dr. Oliver Mothes presents the open-source framework RAGAS, which offers various metrics for evaluating chatbots that can be used not only for quality monitoring but also for optimizing chatbot hyperparameters.

Most entrepreneurs have opened ChatGPT. However, for their own business, it rarely becomes more than a nice conversation. The reason is almost always the same: the transition from idea to application is missing. Where do I start? With what? And is it even worthwhile for a business like mine? We show exactly this transition live in the room. We at Batix know this setting from years of project work. An Excel spreadsheet, a process that eats up time every month, and the question of how a software project even emerges from this. We bring this experience into the workshop. We know how a prototype can be created in the shortest possible time because we do it continuously in real customer projects. In the workshop, we turn this around. The participants bring their own problem – a spreadsheet, a repetitive manual step, an annoying process. Together, we select one or two examples and build a first functional prototype on the spot. At the end of the 90 minutes, we look at a tool that actually runs and can be tested. In doing so, the guests experience something that no lecture can convey. It is about how quickly a real, everyday problem becomes a real, functioning solution. And how little is needed at the beginning.