EDRA: A hybrid architecture for scalable and real-time AI applications

Scalability, low latency, and fault tolerance are critical requirements for modern software systems, especially for AI-powered applications like chatbots and real-time fraud detection. Modern systems often fail to balance real-time processing with scalability, leading to bottlenecks and poor fault tolerance. Traditional architectures, such as Request-Driven (RD) and Event-Driven (ED) models, provide limited scalability and fault tolerance under dynamic workloads. To address these limitations, this paper evaluates a widely used hybrid model, Event-Driven Request Architecture (EDRA), that integrates the synchronous response handling of RD with the asynchronous scalability of ED. This study uses Python’s SimPy framework to simulate EDRA and analyze its performance under varying workloads and failure scenarios. Metrics such as throughput, resource utilization, and latency are measured. Results demonstrate that EDRA achieves approximately 20 tasks per second with 40–50% resource utilization, outperforming RD systems and matching the scalability of ED systems. Additionally, EDRA handles failures effectively using a retry mechanism, ensuring reliability for up to 8 consumers. Our findings highlight that EDRA balances real-time responsiveness with scalable background processing, making it a viable architecture for AI-driven and high-load applications. This study provides a scalable framework applicable to AI systems, e-commerce platforms, and IoT devices. It also offers actionable insights for developers and system architects designing scalable, fault-tolerant systems.