The path towards nearly-Zero Energy Buildings has enforced stricter constraints in construction design while promoting the investigation of new architectural solutions, in residential and producing sectors. Energy simulations, integrated with machine learning, helps academics and professionals to investigate novel strategies for energy saving. We present here a 2-step methodology based on genetic algorithms, aiming to reduce the energy consumption for indoor heating and cooling, while identifying the most suitable commercial solutions for external wall and roof constructions. We compare it with a 1-step optimization algorithm with the goal to determine pros and cons of both methodologies. Even if the two methodologies are comparable in terms of energy reduction, the 2-step algorithm is less computationally expensive and finds several plausible architectural solutions, with equivalent energy profile.