Friction Stir Welding (FSW) is an advanced manufacturing process that joins materials in the solid state using a rotating, non-consumable tool. Accurate and fast prediction of FSW outcomes is challenging due to the strongly coupled thermo-mechanical interactions. In this work, we present a comprehensive computational strategy for FSW, based on high-fidelity, thermo-mechanically coupled Finite Element (FE) simulations, constituting a Full-Order Model (FOM). The simulations compute time-resolved temperature, pressure, and material velocity fields, accounting for heat generation from both, friction at the tool-workpiece interface and plastic deformation within the material. Temperature-dependent viscoplastic constitutive laws, enhanced friction models, and the capability of including tilted and non-axisymmetric tool geometries, enable the reproduction of realistic process behavior across broad parameter windows. [1] Subsequently, snapshots of the FOM outputs are collected into a high-dimensional dataset and reduced via Proper Orthogonal Decomposition (POD) to construct a compact representation that preserves the main thermo-mechanical features of the process. Radial Basis Functions (RBFs) are employed to evaluate the POD-coefficients over a predefined parameter space, allowing rapid evaluation of process responses. [2] The integrated workflow supports efficient FSW process analysis and parameter optimization by making it possible to predict process outcomes orders of magnitude faster than traditional FOM simulations while maintaining high agreement with benchmark results. The proposed framework greatly enhances the capability to explore and optimize FSW process windows through ROMs, offering a robust tool for scientific investigation and industrial process design in FSW.