This comprehensive technical report documents a multi-component approach to in-situ stress characterization at the Utah FORGE EGS site that integrates Machine Learning (ML) methods for predicting near-well principal stresses around geothermal wells with the physics-based finite element model for translating near-field stresses to far-field principal stresses. The ML framework leverages laboratory triaxial ultrasonic velocity (TUV) measurements and field sonic log data to establish velocity-to-stress relationships and estimate the three near-field principal stresses. The physics-based model accounts for thermo-poro-mechanical effects induced by drilling, fluid circulation, and logging operations, as well as stress perturbations associated with the inclined well trajectory. By integrating data-driven ML predictions with physics-based thermo-poro-mechanical modeling, this workflow reconciles near-wellbore stress measurements with far-field in-situ stresses in a geothermal reservoir. Application to FORGE wells demonstrates that near-wellbore thermal and poroelastic disturbances can significantly modify local stress states and that the resulting stress anisotropy is strongly dependent on well orientation. The combined approach provides a robust framework for in-situ stress estimation in complex EGS settings and supports improved interpretation of sonic logs and stress-informed geothermal reservoir development.