This work outlines a Reinforcement Learning framework for dynamic resource allocation in Massive Machine-Type Communications (mMTC) networks, such as IoT systems. The proposed model optimizes bandwidth and power allocation across numerous devices while meeting constraints like energy efficiency and low latency. The RL agent, representing the base station, learns to allocate resources by interacting with the environment and adjusting its decisions based on factors like channel state information and previous outcomes. The system uses deep reinforcement learning techniques (e.g., Deep Q-Learning) to maximize throughput while minimizing power consumption. Several reward-shaping techniques are introduced to refine the agent’s learning, including adaptive penalties and negative reward discounting. This paper also presents performance evaluation metrics, such as average throughput, power efficiency, and fairness, to assess the effectiveness of the model.