This paper presents a modeling framework to enable the creation of custom virtual observer controllers (VO/Cs) to help naval students meet training objectives. First, we describe an approach to create behavior models to drive the behavior of automated entities used in synthetic training. Next, we describe an approach to create behavior models supporting the functions of intelligent VO/Cs. Specifically, we describe the development of an authoring console and training task blocks used to create custom VO/C architectures as well as the creation of student cognitive process models representative of varying training proficiency levels. We also describe a method to integrate these models into simulations used in synthetic training and a communication architecture supporting communication between VO/Cs and our behavior-driven automated entities. We end with a use case of a custom VO/C running end-to-end in the Joint Semi-Automated Forcesâ„¢ (JSAF) simulation in a Navy Anti-Submarine Warfare (ASW) training scenario. Although the Navy ASW training domain is the focus of the examples described in this paper, the modeling framework described is not domain specific.