IfcLLM: Natural Language Querying of IFC Models through Complementary Relational and Graph Representations

IfcLLM is a framework designed for natural language querying of Industry Foundation Classes (IFC) models, integrating both relational and graph representations to optimize attribute retrieval and spatial reasoning. The model achieves first-attempt accuracy ranging from 93.3% to 100% across 30 query scenarios and employs an iterative retry-and-refine reasoning process to handle query failures autonomously. This approach allows for local deployment of an open-weight LLM, making it suitable for data-sensitive architecture, engineering, and construction (AEC) environments, enhancing accessibility to complex building information without the need for specialist knowledge.