Buildings are complex dynamic systems, and traditional HVAC (Heating Ventilation and AirConditioning) control methods, which rely on rule-based algorithms, often struggle to adapt, resulting in inefficiencies and suboptimal indoor conditions. Maintaining and updating these rule-based systems is challenging, as evolving environments require frequent modifications, increasing complexity, and the risk of inconsistencies. Reinforcement Learning (RL) can effectively optimize complex systems like building energy management by adapting through environmental interactions. This study, for the first time, introduces a Deep Reinforcement Learning (DRL)-based HVAC controller named DIET (Dynamic Indoor Environment) using the DDPG (Deep Deterministic Policy Gradient) algorithm to optimize energy efficiency, thermal comfort, and dynamic temperature exposure in buildings. The multi-objective approach of the DIET controller increases complexity but aims to enhance longterm occupant benefits by increasing their exposure to the dynamic variability of indoor temperatures. Developed through framework formulation, online training, and experimental validation, the DIET controller was first trained in the simulation environment of EnergyPlus and later tested in a climatic chamber for heating cases, reducing heating energy use by 28-64 % while maintaining a dynamic indoor environment for 96 % of occupied hours. Results confirm its adaptability and effectiveness, highlighting the potential of the DRL-based controller in HVAC applications. This study addresses the gap in the real-world implementation of smart learning controllers for building applications, provides empirical evidence of the DDPG-based controller's effectiveness, and highlights the need for future advancements in sample-efficient learning and long-term evaluations to improve human-centric climate control systems.