Compliance monitoring of industrial safety remains a major challenge, due to the inherent flaws in single-modality schemes of implementing the highly multidimensional characteristics of workplace hazards. This paper aims to implement a multi-modal deep-learning system that can detect and identify both personal protective equipment (PPE) and facial features and classify and identify the trade to automatically check compliance with safety standards in the industry. The framework uses Detectron2 to get 82.33% on instance segmentation and bounding-box regression, 95.35% on facial recognition with pre-trained embeddings, 92% on a trade classifier based on a convolutional network, and 99.5% on tool detection with YOLO. These modules are merged to form the combined pipeline, which verifies the identity of the workers, verifies certified trade and ensures the use of PPE in real time and then overlaid the safety-status annotations upon visual results. The quantitative results show significant advantage over the existing techniques, such as the capacity to surpass the Average Precision of YOLOv3 by 72.3% in PPE detection. The work is innovative as it includes the multi-modal data in a holistic manner and effectively deals with the interdependencies of the determinants of safety compliance better than the single module models. This study, therefore, offers a scalable, optimized, data-driven model that provided the incidence of workplace accidents through automated surveillance, and where the operability of the system is ensured, by qualitative analyses in which the system is stable in identifying non-compliance situations such as a missing mask or vest, and in all modalities, the model is highly accurate.