Large Language Model–Based Construction Site Management for Severe Weather Preparedness

Severe weather events pose significant risks to construction sites, impacting materials, equipment, and temporary structures. Current preparedness resources from federal and state agencies primarily offer general emergency guidelines, leaving construction practitioners with limited tools for site-specific severe weather preparedness. This study addresses this gap by proposing a large language model (LLM) framework tailored for construction site management, designed to provide proactive recommendations for securing construction materials and incomplete or temporary structures against severe weather impacts. The framework includes three core modules: (1) comprehensive search, (2) data preparation, and (3) LLM development. In the comprehensive search, this study investigated best management practices through an extensive literature review. Next, for data preparation, we augmented text-based raw data using ChatGPT to enhance the diversity of the data set and improve the model’s ability to provide tailored proactive recommendations. Lastly, this study fine-tuned existing LLMs on the customized data set to develop tailored LLMs. A case study demonstrates the fine-tuned models’ effectiveness in generating proactive recommendations, and the performance evaluation results show that the fine-tuned models—particularly the construction material and incomplete and temporary structure 3-30E model (CMITS3-30E Model)—achieved a Bidirectional Encoder Representations from Transformers (BERT) score (BERTScore) F1 of 0.852, delivering optimized proactive actions while balancing computational efficiency. The developed LLM will guide practitioners with precise proactive actions, helping reduce property damage, financial losses, and safety hazards. Overall, this framework not only contributes to severe weather preparedness but also enhances construction safety management by offering tailored, on-demand guidance for construction site resilience.