Hydrodynamic hull-form optimization is a computationally intensive and time-consuming task. It is inherently multidisciplinary, requiring the simultaneous consideration of resistance, propulsion, seakeeping, and maneuvering performance [1][2]. Even in the single-objective case, where only calm-water resistance is addressed, the use of fine computational grids in viscous simulations is prohibitively expensive for integration into optimization frameworks. This limitation persists even when surrogate or approximation models are employed to reduce the number of high-fidelity evaluations [3]. Conversely, the use of coarse grids to decrease computational cost may reduce the sensitivity of the objective function to design variables, potentially misleading the optimization process and degrading its fidelity. To address these challenges, a mixed methodology for hull-form optimization with respect to calm-water resistance, combining potential-flow solvers with high-fidelity viscous simulations is proposed. Potential-flow codes utilizing boundary elements can efficiently capture the influence of geometric variations on hydrodynamic performance over most regions of a ship hull. However, their accuracy deteriorates in the stern region, where viscous effects dominate and cannot be adequately modelled using potential-flow assumptions alone. Based on these considerations, an Artificial Neural Network is developed to account the influence of stern geometry. The ANN is trained to predict the discrepancy between calm-water resistance estimates obtained from viscous simulations and those computed using potential-flow methods, with both approaches employing refined grids. The proposed framework retains the efficiency of potential-flow solvers while compensating for their limitations in viscous-dominated regions through data-driven correction. The ONR Tumblehome model is used as a test case to demonstrate the proposed methodology. The major contribution of this work is that viscous flow calculations are required only during the training phase of the ANN, using the parent hull form and a limited number of variants. The bulk of the evaluations are performed using potential-flow solvers augmented by the ANN correction model.