Heating, ventilation, and air conditioning (HVAC) systems are a major contributor to global building energy consumption; however, their control is complicated by inherent parametric uncertainties and time-varying disturbances. To address the limitations of conventional methods, a novel two-stage "learning-and-reusing" framework was proposed, which fundamentally differs from existing methods by: (i) decoupling parameter learning from disturbance rejection to avoid the single-stage trade-off; (ii) using concurrent learning and an estimator for parameter identification under relaxed excitation conditions and input saturation. In the first learning stage, a concurrent-learning-based adaptive controller accurately identifies key thermodynamic parameters, such as the heat transfer coefficient and the cross-sectional areas of the wall, while simultaneously maintaining precise temperature regulation, thereby building a reliable knowledge base. In the second reusing stage, the identified model is used within a disturbance observer-based robust controller to precisely compensate for time-varying disturbances, such as fluctuating solar radiation and internal heat loads. Simulations on multi-zone building models validated the framework, demonstrating successful parameter convergence and superior robust tracking performance compared to conventional methods. This work offers an efficient, bio-inspired solution for intelligent building thermal management.