One of the most challenging aspects of safety for construction sites is ensuring that workers can predict, identify, and respond to potential hazardous conditions before they are exposed. From a scientific standpoint, we currently lack the knowledge of discovering the most efficient training styles for safety and also understanding why and how these styles of training can influence the post-training activities. To address these needs, a System for Augmented Virtuality Environment Safety (SAVES) is designed and presented in this paper. SAVES, which integrates a BIM with 2D images on a jobsite, allows trainees to conduct a set of interactions with SAVES and accomplish multiple instruction-based and task-based training modules. These modules include detection of ten types of hazard and/or energy sources at three levels of severity. The complete process of design, development and implementation of SAVES is demonstrated and the lessons learned are discussed in detail.