Urban traffic congestion is one of the biggest challenges, in smart city development, impacting travel duration, energy consumption, and air quality in cities. This chapter presents a reinforcement learning based framework for adaptive traffic signal control using Graph Neural Networks (GNNs). Here, we have designed a custom Gym-compatible simulation environment that models a grid, similar to that of urban intersections, where each node represents a traffic signal on the road, managing queues of vehicles. A GNN-based policy network captures the spatial dependencies between intersections and informs a reinforcement learning agent trained using Proximal Policy Optimization (PPO). The model learns optimization of signal phases based on real-time queue states, significantly reducing average queue lengths. All experiments were conducted in Google Colab using Python libraries making the approach both scalable and reproducible. Training behaviour is monitored using through graphs and labelled heatmaps using TensorBoard, and system performance is evaluated through visualizations of reward progression, queue density, and average traffic cumulation. We aspire that this solution demonstrates scalability and robustness for real-world deployment in smart urban mobility systems, directly supporting the Sustainable Development Goals (SDGs) related to sustainable cities and climate action. The approach promotes a very cost effective and data driven resource allocation that has immense possibility to guide the future investments in terms of financials and manpower for urban mobility infrastructure.