PROJECT ABSTRACT This project is about using multiple Graph Neural Network(GNN) architectures as differentiable simulators for running scientific experiments in fluid mechanics and/or electromagnetism domains. Furthermore, assuming that the simulators are trained well enough so that they learned the underlying physics, the end goal is to use these simulators for designing engineering tools that may be of use in the aforementioned domains. These tools can be simple POC prototypes, AI-designed airfoils, or antennas. The project's goal is to start with tools being able to do simple manipulation tasks and progressively increase the complexity to tools that may have a potential to be manufactured. Lastly, the project also aims investigate the deficiencies of different GNN architectures in terms of their potential to capture long-range interactions between different nodes in their training set. It is known that these architectures are excessively prone to vanishing gradients. There have been several remedies to fix this inherent disadvantage, however, those remedies have not been compared against each other in real world tasks of interest. We hope to see the advantages/disadvantages of different architectures in engineering design tasks.