We present Levy-scORD, a mechanistic, interpretable framework for prediction and control of cell-state trajectories in glioblastoma (GBM). Levy-scORD models transcription-factor and gene-expression dynamics using L´evy-driven stochastic differential equations with compound Poisson transcriptional bursts and state-dependent discrete jumps, thereby capturing bursty gene regulation, heavy-tailed fluctuations, and rare phenotype transitions. By integrating high-throughput single-cell RNA sequence data with a stochastic optimal-control formulation, Levy-scORD enables principled design of personalized multi-drug strategies that minimize expected drift toward resistant phenotypes. We test our model on glioblastoma scRNA-seq datasets, where it yields interpretable regulatory parameters, and provides an effective way to identify potential drug targets to prevent therapy evasion. Furthermore, the framework also serves as a general tool for synthetic data generation and forecasting of single-cell dynamics GBMs across different patient types.