Weaving Knowledge Graphs and Large Language Models (LLMs): Leveraging Semantics for Contextualized Design Knowledge Retrieval

Demographic change in Europe challenges companies as retiring employees take valuable expertise with them. To address this, knowledge graphs (KGs) are emerging as tools for structured knowledge representation. Simultaneously, large language models (LLMs) are increasingly being used as innovative solutions for information retrieval. However, LLMs generally process only public knowledge, and recent approaches integrating Retrieval Augmented Generation (RAG) for private knowledge retrieval often lack contextual relevance. To enhance trustworthiness and overcome these limitations, a method is proposed for embedding latent problem-solving structures within design processes into LLM-driven information retrieval systems. Using a case study in energy infrastructure, a KG of design problems was constructed by extracting functional requirements from semi-structured documentation via LLMs. This KG is further utilized by an LLM to answer queries, with results visualized through an interactive interface. Validation through field studies with engineers underscores the approach’s effectiveness in enhancing contextual and trustworthy knowledge dissemination.