Ensuring the safety of workers in manufacturing and construction, and responding quickly to incident events, has become a pressing challenge. Traditional CCTV monitoring has limitations, as the number of the cameras are bound to the number of personnel for visual surveillance. And while conventional machine learning algorithms can analyze videos, they often require extensive training on specific event types. This presentation will demonstrate a system that leverages AWS Cloud services to address these challenges. By feeding video stream into a generative AI model, the system can analyze the captured situations in real-time and provide alerts. Unlike conventional AI, this approach does not need to be trained on each type of incident. Instead, it can adaptively adjust its behavior to flexibly detect dangerous situations like accidents and injuries. This helps reduce the manpower required for constant monitoring, while enabling swift response.