This study presents an AI-BIM integrated construction safety management and educational training system that leverages computer vision, deep learning, and virtual reality technologies to enhance both on-site safety supervision and engineering pedagogy. Traditional safety supervision methods lack efficiency, adaptability, and educational value. This study aims to address these limitations by developing an AI-BIM integrated system that consists of two core components: real-time safety behavior detection and immersive safety training. Using OpenCV and YOLOv3, the system automatically detects the use of personal protective equipment (PPE) such as safety helmets and safety vests. Individuals not complying with safety regulations are identified via facial recognition, followed by automated SMS alerts. Repeated violations trigger a mandatory re-education process involving VR-based simulations built upon BIM data and a personalized safety test. Experimental results demonstrate that YOLO significantly outperforms traditional OpenCV in detection accuracy, particularly under complex construction site conditions. Beyond technical improvements, this platform serves as a teaching tool, allowing engineering students to experience intelligent safety systems through simulation. They engage in case-based analysis, improve model accuracy, and internalize safety responsibilities through hands-on experience. The proposed system thus contributes to both smart construction site management and innovation in engineering education.