Discrete Optimization via simulation (DOvS) studies the optimization of complex systems with discrete-valued decision variables using simulation. This talk will present my research on the development of "industrial strength" DOvS for realistic scientific and engineering applications where simulation is time-consuming and fraught with data and model uncertainty. To efficiently solve DOvS with time-consuming high-fidelity simulations, we propose a new multi-fidelity optimization framework known as MOTOS. MOTOS uses low-fidelity simulations to broadly explore the solution space and focuses high-fidelity simulations on promising solutions. MOTOS integrates low- and high-fidelity simulations in an efficient and rigorous manner and has been shown to lead to significant computational savings. To quantify and mitigate the impact of uncertainty in DOvS due to data variability and model mis-specification, we propose a disparity-based data-driven approach that provides both asymptotic efficiency and distributional robustness. We establish the convergence properties of a sample average approximation scheme, and demonstrate the efficiency and robustness of the new approach using two illustrative problems. Finally, I provide a brief overview of Industrial Strength COMPASS (ISC), an open source DOvS solver written in C++. ISC has been shown to outperform a leading commercial DOvS solver on challenging DOvS problems, and it guarantees convergence and valid statistical inferences that commercial solvers lack.