As urbanization increases around the world, high-rise buildings will continue to become a more prevailing typology, nonetheless, due in part to cumbersome computational simulations, rarely do designers have enough information during the early stages of design, which is the time when their choices affect the most the efficiency of their building. Surrogate models, aka meta-models that predict how the original simulation models behave offer a clear advantage in terms of speed of the results. This study delves into performance-based design using surrogate models to give the designer a tool to quickly understand the variables that will affect its efficiency. Looking specifically to improve four (4) results: energy consumption, natural daylight, comfort, and floor area. This study contemplates 16 unique variables ranging from effects of the Context (1), general building shape & orientation (6) to façade variables (9). The energy results are validated in DesignBuilder software before proceeding to collect 500 samples for two different locations: Bogotá and Amsterdam. This data is then run through three machine learning methods, Multilinear Regression, Non-linear Regression, and ANN. Next, the chosen ANN-based surrogate models for each of the outcomes are trained and hyperparameters finetuned to increase their R2 value and reduce their standard error (MSE) and mean absolute error (MAE). Finally, the generic surrogate models are run and compared through various optimization algorithms to determine Pareto-frontier options that ultimately improve the energy performance of a solution with the daylight, comfort, and floor area as design constraints or goals. A time reduction of up to 99.96% was achieved to collect another 500 samples. Finally, the final model also serves as an aid for visualization of the design space by allowing near-real-time (6 seconds) to generate the form of each design solution