A deeper theoretical understanding of electrified interfaces becomes important for the development and control of electrochemical devices. While much progress has been in computational electrochemistry, constant potential simulations pose fundamental challenges. Here, we report a multiscale quantum mechanical / machine learning (QM/ML) approach, effectively reducing the computational time, but maintaining the accuracy of QM with an explicit description of the electrolyte components. Implementation of a bias potential to the electrodes, maintained with a bespoke computational potentiostat, allows the system to achieve the constant potential (uVT) ensemble. Charging dynamics and electrolyte response are analyzed as a function of the electrode potential for water molecules incorporated in bilayer graphene. Multiscale embedding theory that takes advantage of increasingly powerful machine learning surrogate models offers a promising avenue to model realistic electrochemical systems.