IFC Whisperer: Querying Building Information Models with Large Language Models and IFC-based Knowledge Graphs

The transition to sustainable and resource-efficient built environments requires systematic access to semantic building information to improve performance management. However, current Building Information Modeling (BIM) practices face challenges in making geometric, topological, and material data easily accessible. Extracting critical information, such as material properties, often requires expert knowledge and complex parsing, limiting the effective use of BIM for material lifecycle management and waste reduction.

To address this challenge, this paper presents a novel approach that leverages graph-based retrieval and large language models (LLMs) to enable intuitive BIM interaction through natural language queries. The proposed pipeline transforms Industry Foundation Classes (IFC) data into a dynamic knowledge graph and then uses LLM to translate queries (for example, ’Which materials in the building can be recycled?’) into structured graph searches.

Comprising three key steps: (1) extraction of entities from IFC models, (2) construction of knowledge graphs, and (3) translation of LLM-supported queries, the framework facilitates more accessible decision-making in sustainable building practices, particularly for material life cycle management and waste reduction.