Achieving multi-objective optimization for building envelopes across diverse climates is a critical yet complex design challenge, essential for mitigating energy costs and climate change. This paper presents an AI-integrated framework to automate the prediction and optimization of residential building performance, validated across the distinct climates of Turin and Naples, Italy. A Python-based architecture synergistic with EnergyPlus simulations was deployed to optimize envelope configurations from a 14,400-combination parametric space (2,880 scenarios sampled via LHS), systematically balancing Energy Use Intensity (EUI), heating/cooling demands, and occupant comfort (PPD, PMV). From a comprehensive suite of eight machine learning models, CatBoost and a Keras Tuner-optimized Artificial Neural Network (KerasTunerANN) emerged as superior for the Turin and Naples climates, respectively, based on their mean error metric ranks (1.40).