This paper describes the implementation of an AI system that can perform the dual role of Job Performance Aid (JPA) and Intelligent Tutor (IT) for use in On-the-Job Training (OJT). It is well known that the best human experts possess a mental model of internal equipment operation and a good trainer will teach this conceptual knowledge as well as the usual diagnostic skills. The Intelligent Tutor portion is aimed at building this mental model through interaction with a simulation of the equipment. The student interface employs high resolution graphics and a mouse. The simulation is a qualitative causal model which is much simpler than a full mathematical model yet retains all the important distinctions between system states. The Job Performance Aid is an Expert System (ES) which is automatically derived from the qualitative simulation model. This is accomplished by using the model to predict the behavior of the equipment and the propagation of effects under all conceivable conditions. The ES rules are then induced from the fault symptom pattern produced by exercising the model. By taking this approach, the ES provides "deep reasoning" as opposed to the "shallow reasoning" often found in an ES based solely on externally observable features.