Current brain-in-the-loop systems often rely on generic parameter configurations that overlook inherent neural variability. This research aims to improve system performance by implementing closed-loop optimization to personalize parameters for each user and adapt them in real-time. Conventional "one-size-fits-all" approaches are suboptimal because they fail to account for inter-individual differences and intra-individual fluctuations (non-stationarity). To address this, we utilize Bayesian Optimization, a sample-efficient method ideally suited for to cope with the noisy, expensive-to-evaluate neural signals. The project pursues four specific objectives: (1) Static Optimization, identifying personalized parameter sets for individual users; (2) Dynamic Optimization, continuously adapting parameters to track changing neural states; (3) Model-Informed Optimization, integrating physiological priors and/or generative surrogate models to guide the search; and (4) Generalization, validating the framework in fundamental neuroscience applications beyond BCI.