<p>Traffic congestion remains a pressing challenge for urban mobility, affecting travel times, safety, and environmental sustainability. This paper presents a practical hybrid optimization framework for traffic signal timing that integrates a YOLO-based vehicle detection module with a SUMO-based traffic simulation and optimization module. The system was developed in collaboration with the National Road Safety Agency (NARSA) in Morocco as part of a larger Intelligent Transportation System initiative. The detection module, built using the YOLO object detection architecture, has been tested on recorded video streams due to current regulatory restrictions on installing live cameras; it is designed to operate identically on live feeds once authorized. The simulation and optimization module allows users to select a modeled intersection within a web application, specify traffic demand and constraints, and run simulations in SUMO to compute optimal traffic signal cycle times using a hybrid Genetic Algorithm (GA) and Stochastic Gradient Descent (SGD) approach. The current study evaluates the optimization framework in simulated conditions for the Bab Zaer intersection in Rabat, Morocco, under varying traffic demands. Results show substantial reductions in total and per-vehicle waiting times compared to baseline configurations. While this work focuses on simulation-based validation, the proposed architecture supports future integration with live detection data and remote programming of traffic controllers, enabling scalable deployment in real-world traffic networks.</p>