Researchers have developed AISA, a framework that uses large language models to improve highway construction safety. The system analyzes incident narratives and classifies them for reporting purposes, while also retrieving relevant historical accidents and industry documents to inform daily safety plans. AISA was tested on over 15,000 narratives and achieved high accuracy in identifying incidents and retrieving relevant information. However, the quality score derived from the model was distorted when applied to out-of-distribution fatalities. --- Why it matters: This matters because it provides a new framework for improving highway construction safety through data-driven decision making. By leveraging large language models, AISA can help reduce accidents and improve worker safety in this high-risk industry. Source: https://arxiv.org/abs/2608.17184