This thesis focuses on the development of a controller for the management of an HVAC (Heating, Ventilation and Air Conditioning) system, in collaboration with DENSO Thermal Systems. The work builds upon an already available neural-network-based surrogate model of the HVAC unit and aims at extracting control-oriented information through a dedicated analysis of the model behaviour and of the admissible operating region. The overall activity is carried out in both MATLAB and Python, leveraging the two environments for model handling, optimization, and validation tasks. The MIMO controller is formulated as an optimization problem in which, given a set of airflow references at the outlets, the objective is to compute an optimal input command vector composed of the individual outlet openings and the blower speed.