A vast number of continuum mechanics problems are modeled using Partial differential equations. Physical and geometric complexities of engineering problems render analytical solutions for most PDEs intractable, leading to reliance on numerical methods. However, most traditional numerical methods rely on accurate domain discretization and expert knowledge to develop consistent, stable numerical schemes. Physics-informed neural networks have recently shown promise to alleviate some of these issues. However, they suffer from long training times, arbitrary hyper-parameter requirements, and inconsistent predictions, making them inadequate for engineering problems. Here, we improve upon the current state of Physics-informed neural networks and provide ways to overcome some issues plaguing their wide application. We apply our methods to solve many problems in solid mechanics involving cracks and material non-homogeneity and problems in fluid mechanics involving boundary layers. We also propose a new way of constructing novel basis functions for PINNs to provide accurate, stable solutions for PDEs.