This paper presents a novel optimization methodology based on both adjoint sensitivity analysis and trust-based dynamic response surface modeling to improve the performance of a modern turbine of a large civil aero-engine in the presence of high-fidelity geometry configurations. The system has been applied to the nonaxisymmetric hub and tip endwall optimization of a high-pressure turbine stage making use of multirow 3D simulations, parametric modeling, and rapid meshing of real geometry features such as rim seals and modeling of film cooling flows. It has been shown in previous papers that improvements gained using simplified models of the stage are lost when applying the high-fidelity geometry configuration. New results presented in this paper indicate that controlling the purge flow that exits the disk space through the rim seal at the hub of the main annulus is more significant than the reduction of secondary flows in the main passage. For a given rim sealing mass flow rate and whirl velocity, the nonaxisymmetric endwalls are optimized such that the detrimental impact of the sealing flow on the turbine performance is reduced, and hence, the stage efficiency is significantly increased. The traditional optimization approaches based on evolutionary methods or even sequential modifications for defining the endwalls shape are computationally demanding. Since turbomachinery industry continuously strive to reduce the design cycle time, in particular when high-fidelity 3D computational fluid dynamics (CFD) is used, the main body of this paper outlines the novel methods developed to produce a practical design in a very aggressively short design cycle time.