The objective of the proposed research is to develop scalable optimization and learning methods that restore AC feasibility in power system operations while preserving the computational advantages of the DC optimal power flow (DCOPF)→AC power flow (ACPF) pipeline. Although DCOPF is widely used in power system operations and planning due to its computational efficiency, its simplified representation of network physics can produce dispatches that violate the nonlinear AC power-flow equations or associated operational limits. To address these limitations, this research develops loss-aware DC formulations, structured AC power-flow restoration methods, and differentiable control policies for distributed slack and PV/PQ regulation. The research further extends these approaches toward end-to-end optimization and security-constrained operation. Together, these methods aim to bridge computationally efficient DC-based optimization with the physical requirements of reliable AC grid operation.