Accurate radiotherapy planning for glioblastoma is limited mainly by two factors: microscopic tumor infiltration that is invisible in routine MRI and the mechanical displacement of brain structures by mass effect, induced by the growing lesion. As a result, standard treatment planning relies on uniform safety margins that ignore patient-specific growth patterns and deformations. This motivates approaches that combine mechanistic priors with data to improve target volume estimates. We present our framework for a hybrid, patient-specific simulation that couples a reaction-diffusion type tumor growth model to a biomechanical description of deformable brain tissue. The forward model simulates tumor cell density evolution together with the deformation field in anatomically realistic settings, accounting for heterogeneous tissue properties and varying diffusion, proliferation, and material parameters. We describe a forward simulation pipeline that generates synthetic tumor growth and mass-effect scenarios on realistic anatomies and model parameters, including controlled parameter sweeps and reproducible synthetic data generation. Building on this forward process, our work targets the inverse problem of personalizing model parameters from clinical imaging. We also consider sensitivity and identifiability as practical constraints on what can be inferred from limited observations. Because classical optimization and Bayesian inference over coupled PDE models can be computationally demanding, we developed learning-based surrogates that approximate the forward solver with low latency. To capture mass-effect properties, a domain-agnostic surrogate is trained on a large synthetic dataset generated by the simulator under controlled variations of anatomy, initial conditions, and parameter distributions. The surrogate is embedded into gradient-based, Bayesian, and evolutionary calibration pipelines, enabling rapid parameter estimation, uncertainty quantification, and probabilistic maps of infiltration and recurrence risk. We discuss modeling choices, data generation strategies, baseline comparisons, and evaluation concepts on synthetic benchmarks, with perspectives toward retrospective clinical studies and integration into radiotherapy planning workflows.