Drilling performance optimization remains one of the most critical factors influencing the success and cost-effectiveness of geothermal and CCS operations. Achieving consistent performance under highly variable subsurface conditions requires accurate, reliable, and adaptive predictive models. Historically, drilling optimization relied heavily on empirical models, statistical regressions, and operator expertise. While these approaches provided insights into key performance metrics, such as ROP and shock dynamics, they often fell short in capturing the nonlinear, multi-dimensional relationships in drilling data. Furthermore, static models based on idealized assumptions struggle to adapt to real-time operational changes, sensor noise, and equipment degradation.