Compliance checking of construction requirements is critical to ensure that the project is completed and commissioned while meeting the code, standards, and owners’ needs. However, traditional compliance-checking approaches struggle to deal with extensive documentation and laborious processes, leading to omissions and inefficiencies. Existing rule-based and machine learning approaches lack adaptability, scalability, and domain-specific understanding. This study addresses these challenges by introducing a generative artificial intelligence (GenAI)-assisted compliance-checking system that consists of three phases: (1) structuring requirement clauses with an ontology and extracting building information modeling (BIM) project data; (2) fine-tuning (FT) a large language model (LLM) with ontology concepts and BIM metadata to improve domain-specific knowledge contextualization; and (3) developing a two-step relevance ranking strategy (RRS) to ensure effective and accurate retrieval of requirement clauses, which will integrate project attributes and generate structured compliance-checking reasoning using a GenAI model. A benchmark data set of 100 scenarios of construction compliance checking and evaluation metrics for generated checking outputs are introduced to assess retrieval accuracy and interpretability of reasoning output. Experimental results show that the proposed approach improves requirement retrieval and reasoning accuracy. This GenAI-assisted system balances automation and expert oversight, allowing construction professionals to efficiently automate compliance checking while retaining control over final compliance decisions. The method, data set, and metrics lay the foundation for a GenAI-assisted compliance-checking system that is scalable, interpretable, and aligned with real-world construction practices.
Wang et al. (Fri,) studied this question.