In this paper a network modeling approach is introduced to integrate multiple important elements for social interaction between a human-like AI Coach and its users in sometimes stressful circumstances. The computational approach integrates concepts from neuroscience such as generating multiple emotions, mirroring and emotion contagion, self-other distinction and emotion attribution, empathic responding, individual emotion regulation, and interpersonal emotion regulation. It is shown how, for example in stressful circumstances, such a human-like AI Coach over some time period is able to (1) resonate with a user’s emotions and show empathy, and (2) if needed regulate the user’s emotions.