In the industrial computational fluid dynamics (CFD) landscape, Reynolds-Averaged Navier-Stokes (RANS) turbulence models constitute the primary tool for aerodynamic design and analysis, owing to their computational efficiency and capacity to deliver physically meaningful predictions. Nevertheless, certain flow configurations, particularly those involving turbulence separation and detachment, exhibit substantial predictive discrepancies when compared with experimental data. While Large Eddy Simulation (LES) approaches offer superior fidelity, their prohibitive computational demands render them impractical for high-Reynolds-number industrial applications. This thesis addresses this limitation through a data-driven calibration frame- work designed to enhance the accuracy of RANS predictions on critical test cases exhibiting significant improvement potential.