06. FVI Experts' Breakfast

Topic: The new colleague is an AI – Large Language Models (LLM) and RAG in maintenance. In this session, Marcel Hahn and Jens Reisenweber left the theoretical ivory tower and demonstrated live how Generative AI (GenAI) revolutionizes the daily work of maintenance personnel.

• The end of searching (RAG technology): The biggest problem is not the lack of knowledge, but finding it again. With "Retrieval Augmented Generation" (RAG), it was demonstrated how to chat with an AI that only accesses your own documents (PDFs, manuals, notes).
• Live proof 1: A defective power supply. Instead of leafing through manuals, you ask the AI: "Which power supply do I need according to the specification?" – The AI immediately provides the type and even writes the order email to the supplier (Weidmüller).
• Breaking barriers (language & format): Jens Reisenweber impressively showed how AI solves language barriers in the skilled labor shortage. A Hungarian shift report (text) was automatically translated, analyzed, and turned into a clean German Excel table for controlling.
• Finding hidden knowledge (saw splitter case): Using a "saw splitter," it was shown how the AI links scattered information. The 20-year service life of a safety component was not explicitly stated as a sentence anywhere but was spread across three documents. The AI still deduced it and issued a warning. This is a feat that a human under time pressure would often overlook.
• Fear of job loss vs. superpower: The discussion (including Heinz-Achim Schulte) revolved around the concern "Will this make me unemployed?" The consensus: No, but it makes you efficient. What used to require 4 people for documentation and research can now be done alone, allowing focus on real problem-solving. AI is not a replacement but an efficiency booster.
• Knowing limits (math & hallucinations): Transparency is important. LLMs are language models, not calculators. With complex mathematics or missing data, they can "hallucinate." Therefore, the human is always needed as the final instance ("Human in the Loop").

Classification: From theory to "Asset Intelligence" This breakfast was the technological proof of everything we discussed in episodes 1–5.

• ADAM is not future music: What was shown here as an "Asset Intelligent" prototype is the core of ADAM. We are not selling a vague vision but a technology that already turns Hungarian voice messages into German maintenance plans today.
• The answer to the skilled labor shortage: If we can no longer find German-speaking engineers, we need tools that understand every language and present knowledge so simply that even trained personnel can maintain complex systems ("power steering for employees").
• Data sovereignty in practice: The demo showed that you can lock the AI in a "closed room" (e.g., Azure). The company's knowledge does not end up in the public ChatGPT training pool but remains in the customer's digital vault. This is the strongest argument against the fear of data leakage.

Conclusion: The maintenance of the future does not type into SAP masks, it talks to its systems. We have shown: The technology is mature. Now it's just about having the courage to use it.