In this proceeding, we introduce physics-driven deep learning for the study of strongly interacting systems. We demonstrate that the integration of deep models with physical principles can lead to more efficient and reliable solutions, learned from both first-principles computations and observational data. Specific applications include hadron-hadron interactions and the reconstruction of neutron star equations of state (EoSs). Deep neural networks (DNNs) are employed to represent physical quantities with known symmetries and rules, trained to optimally describe observational data.