LLM.assist: Customizable, hardware-efficient on-premises AI assistance for domain-specific needs—supporting SMEs from technical problem solving to tactical and strategic decision-making

Especially for SMEs, the requirements arising from connectivity, regulatory complexity, economic viability, sustainability, and internationalization—and their many implications—are becoming increasingly difficult to manage. Here, artificial intelligence offers tremendous potential to analyze trends, patterns, and relationships in large-scale, heterogeneous data.

The project aims to develop domain-specific and individually customizable AI assistance based on Large Language Models (LLMs). It is intended to support questions ranging from technical topics (e.g., production and maintenance) to strategic issues (e.g., business intelligence) and to present validated results in an intuitive way. The R&D focus areas include LLM tailoring/tuning, a dedicated intermediate representation (intermediate language), adaptable semantic models, and feedback-based validation of results against the knowledge base and domain-specific criteria, as well as powerful visualization and use-case-specific customization for different target groups (engineering and management). Designed as a hardware-efficient on-premises solution, the technology offers significant economic exploitation potential and makes an important contribution to increasing the efficiency and competitiveness of SMEs.

Subproject: Research into a flexible, hardware-efficient approach for application-specific adaptation, energy-/performance-efficient configuration, and targeted use of Large Language Models as AI assistance for augmented analytics