semio : Large-language-model-based building-information-model alignment for automatic-compliance-checking - towards closing the gap between model authoring and model checking for kit-of-parts architecture

Saluz, Ueli and Baimuratov, Ildar and Geyer, Philipp; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) semio : Large-language-model-based building-information-model alignment for automatic-compliance-checking - towards closing the gap between model authoring and model checking for kit-of-parts architecture. In: EG-ICE 2025. University of Strathclyde Publishing, GBR, pp. 272-281. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093288)

[thumbnail of Saluz-etal-EG-ICE-2025-semio-Large-language-model-based-building-information]
Preview
Text. Filename: Saluz-etal-EG-ICE-2025-semio-Large-language-model-based-building-information.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo

Download (1MB)| Preview

Abstract

Designing regulation compliant buildings is challenging. An iterative design process involving architects and experts is vital. On the design side, many authoring frameworks have been developed to enable a flexible design process. On the constraint side, many rule-based frameworks have been developed for Automated-Compliance-Checking (ACC). A general explicit algorithmic translation from authoring to constraint formats is rarely possible. An additional framework expert is needed, which limits ACC. Closing the gap between model authoring and model checking with automated model alignment is crucial. While black-box Large-Language-Models (LLMs) struggle with knowledge-intensive tasks, they are powerful in loosely instructed conversion tasks. The opposite holds for white-box symbolic reasoning frameworks such as Web-Ontology-Language (OWL). We propose to use a Chain-of-Thought (CoT) based model alignment from authoring to constraint formats. Minimizing the Framework-Expert-In-The-Loop. A Kits-of-Parts (KoP) test-case with fire safety constraints demonstrates that LLMs are capable of reasoning to align designs with regulations, enabling ACC.