From research prototype to industrial knowledge extraction

Unstructured data sources such as reports, emails and tacit expert knowledge present industrial companies with the challenge of efficiently extracting and utilizing relevant information for decision-making processes. Conventional approaches reach their limits here: they are unable to fully capture the complexity and diversity of the data.

How can tacit experiential knowledge in industry be captured and made usable with artificial intelligence? This question was at the heart of the PACT INSIGHT research project, which the Cognitive Service Systems Research and Innovation Center KODIS at Fraunhofer IAO carried out in collaboration with Fraunhofer Austria. The project entered its next phase with the TACIT-Talk initiative. Together with semiconductor manufacturer Infineon, the approach was then further developed in industrial practice.

In the previous »PACT INSIGHT« project, a demonstrator was developed that makes implicit experiential knowledge visible through the combined use of large language models (LLMs) and knowledge graphs. Through an intuitive voice interface, experts can input their knowledge; the system automatically transcribes and structures the content and links this information semantically. This has resulted in an innovative approach to AI-supported knowledge management that makes experiential knowledge available for the long term.

Based on the findings from the project, the follow-up project »TACIT-Talk« was launched. The project applied the approach developed in »PACT INSIGHT« to real-world industrial scenarios. The goal was to capture and make experiential knowledge accessible on a large scale, particularly in industrial maintenance. To achieve this, LLMs and knowledge graphs were combined to link unstructured and structured data—from maintenance logs to expert notes. A key milestone: the collaboration with Infineon.

The semiconductor manufacturer tested »TACIT-Talk« in a production environment—a testament to the industry’s confidence in the developed approaches. The collaboration kicked off at the project launch in Villach in July 2025. Here, researchers tested how AI-supported knowledge capture can increase the efficiency and quality of maintenance processes.

Project partners:

• Fraunhofer Austria Research GmbH (FhA)
• Infineon Technologies Austria AG