The exponential growth of video data from surveillance cameras and IoT devices has highlighted the need for real-time analytics that minimize resource consumption while maintaining high accuracy and efficiency. This paper introduces a scalable and efficient architecture designed to process large-scale video streams with zero data loss, leveraging Apache Kafka, Apache Spark Structured Streaming, and YOLOv8 for dynamic sub-stream generation and real-time object detection. By utilizing Spark’s concurrent data processing capabilities, the system optimizes resource usage, reducing the demand for CPU, RAM, and network bandwidth while delivering fast and accurate results. Additionally, Apache Airflow orchestrates workflows across distributed microservices to ensure seamless operations. Preliminary findings validate the architecture’s ability to handle high-throughput workloads effectively, with future improvements focusing on Kubernetes-based deployments and advanced filtering mechanisms.