Reincarnate is pleased to share the publication of the paper IFC Whisperer: Querying Building Information Models with Large Language Models and IFC-based Knowledge Graphs,published in the Journal of Physics: Conference Series as part of CISBAT 2025.
Authored by Yu Hsiu Tung, Samaneh Rezvani, Fatemeh Asgharzadeh, Maurijn Neumann, and Timo Hartmann, the paper presents a new approach for making BIM data more accessible by combining IFC knowledge graphs with LLM-supported natural language querying. The method transforms IFC models into a structured graph representation and enables users to retrieve material, geometric, and lifecycle information through intuitive questions rather than complex IFC parsing.
The framework consists of three components: semantic enrichment of IFC to extend sustainability-related attributes; graph-based representation to preserve relationships between materials, components, and property sets; and LLM-based query translation to allow users to retrieve information using natural language. The study shows how this combined process enables practitioners to explore properties such as recycled content, volumes, manufacturer-specific attributes, and other sustainability parameters in a more direct and explainable way.
To demonstrate the approach, the authors applied the framework to a case study on Digital Delivery Ticket (DDT) data for concrete. The enriched IFC model included attributes such as GWP values, recycled aggregate percentages, and manufacturer data, all of which were mapped into a knowledge graph and queried using LLM-generated Cypher statements. The results showed that graph-based querying significantly outperformed direct LLM interpretation of IFC text, especially for multi-hop queries that require navigating relationships across building elements.
This work supports Reincarnate’s mission to improve digital workflows and unlock the potential of data-driven circularity across the built environment.