Full-wave electromagnetic simulation of reconfigurable terahertz (THz) nanophotonic devices entails severe computational expense, rendering conventional brute-force parametric sweeps and evolutionary metaheuristic optimization algorithms (such as GA and PSO) computationally prohibitive. In this paper, we present an autonomous, sample-efficient closed-loop Bayesian Optimization (BO) framework for dynamic programmatic tuning of graphene's chemical potential (μc) for optimal impedance matching in terahertz patch antennas. The electromagnetic model comprises a single-layer graphene radiating patch on an engineered High-Resistivity Silicon / Polyimide double-stack substrate resonating at 1.5 THz, whose complex surface conductivity is governed by the quantum-mechanical Kubo formulation. Treating the full-wave Finite Element Method (FEM) solver Ansys HFSS as an expensive black-box objective function, we formulate a Gaussian Process Regression (GPR) surrogate model equipped with a Radial Basis Function (RBF) covariance kernel and an analytical Expected Improvement (EI) acquisition function. The algorithm dynamically explores and exploits the continuous Fermi-level space μc ∈ [0.15 eV, 0.95 eV] by programmatically re-evaluating the complex surface impedance (Zs = Rs + jXs), updating simulation design variables, triggering adaptive meshing solutions, and extracting return loss (S11) metrics without human intervention. The proposed Bayesian framework achieves global convergence to the optimal chemical potential (μc* = 0.542 eV, yielding S11 = -27.42 dB and VSWR = 1.09) within only eight solver iterations, saving greater than 90% in computational time compared to standard Genetic Algorithms (GA) and Particle Swarm Optimization (PSO). The framework is further benchmarked through comparative evaluations against Lower Confidence Bound (LCB) and Probability of Improvement (PI) acquisition functions, sensitivity audits across kernel hyperparameters, and validation of numerical stability in automated script-driven terahertz nanophotonics.