We are developing data-driven models of neural systems (from neural populations to whole brain) to understand and predict stimulus-invoked neural and behavioral responses. We utilize deep generative modeling techniques (such as variational inference and Flow matching) along with dynamical systems to build these models. The goal of this work is to investigate these models to gain understanding about the neural computations underlying behaviour through the lens of dynamical systems theory, as well as to potential leverage these models for model-based neural stimulation.