We develop Physics informed Machine Learning architectures that provide a faster way to predict solutions to wave physics problems. Classical methods, such as FEM/FDTD, may be computationally very expensive or even unfeasible on some cases. Purely data-driven Machine Learning struggles with spectral bias, generalization and low data regimes. Finally, inverse problems are ill posed and undetermined and need very complicated computations to perform with classical methods. We aim to produce architectures that can deal with all these problems by integrating physics priors into the learning process.