This project aims to develop a demand-driven approach for shared mobility operations with machine learning and math programming methods. The objective of this approach is to incorporate economic, environment and equity impacts over an entire operational cycle. Both ride-hailing systems (e.g. Lyft) and ride-pooling systems (e.g. UberPool) will be investigated. The developed models are tested with real-world taxi data including detailed trajectories of vehicles and their loading states at all times. We proposed a deep Q learning model to optimize the system performance. In the model, we train a Q network and operate the system with real-time demands. Services include serving travel demands, rebalance, charging in charging stations. Case study in NYC is explored and the results are analyzed in the cast study section. Further, we propose a mathematical programming approach for solving the vehicle routing problem for a high-capacity ride-pooling system. This approach attempts to extend the existing Request-Trip-Vehicle-graph based method that used heuristic methods to generate vehicle routes into one that use customized column generation to generate candidate optimal vehicle routes more efficiently.