Predictive behavior modeling poses several difficult challenges. Human behavior modeling using rational choice theory, negotiation protocols, and other socio-economic models have been somewhat successful in prediction of events on a city-wide or state-wide level (or a combination of levels) although tactical level simulations often do not consider this level of complex human interplay. Often seemingly small or insignificant tactical level events have lead to socio-political situations that shape the course of wars, influence political and policy changes, create areas of hostile incubation, and affect economic and social climates. We propose an immersive tactical level simulation framework that provides a novel method of modeling social complexity in which virtual agents perceive events and share their interpretations of events. The framework uses an open source technology design with an emphasis on generating extensible agent interaction models and realistic representations of agent's actions, gestures, communications, and responses in a virtual training environment. The organizational dynamics generated by the modeling approach produce a highly variable set of possible outcomes to the training scenario. Combined with specific learning objectives, this high degree of variability within a learning environment poses new challenges to the trainee, namely the need to be aware of how to operate in highly dynamic environments. We propose a model for simulating aspects of social complexity using an agent-based immersive training system and describe how these techniques can be applied to the development of cross-cultural competence, situational awareness, and crowd behavior analysis.