Lorenzo is developing a tool called Safety AI for DroneDeploy, aimed at enhancing construction site safety through advanced image analysis. Launched in October 2024, it utilizes “reality capture” imagery to identify conditions violating OSHA regulations with 95% accuracy. Unlike similar AI tools that rely on human workers for object labeling, Safety AI employs generative AI, allowing it to interpret complex site images and generate conclusions regarding safety violations. This tool is being implemented across hundreds of U.S. construction sites, with adaptations for regulations in Canada, the UK, South Korea, and Australia. Lorenzo’s approach includes creating a comprehensive “golden data set” of OSHA violations to train the model, ensuring accuracy through ongoing interaction with construction safety professionals. The visual language model (VLM), a blend of a large language model and visual analysis, enhances the AI’s problem-solving capabilities, requiring human oversight to refine its outputs and ensure precision in identifying safety risks.
Source link
