An Empirical Evaluation of an Event-Driven Modular Monolithic Architecture Using Apache Kafka for Real-Time AI-Powered Systems

Recent academic and industrial research highlights a persistent architectural dilemma between monolithic and microservices-based systems. While microservices architectures provide scalability and independent deployment, multiple empirical studies report substantial increases in operational complexity, recovery overhead, and maintenance cost. Conversely, monolithic systems offer simplicity but suffer from synchronous communication bottlenecks and limited fault isolation. This paper presents a comprehensive empirical evaluation of an Event-Driven Modular Monolithic Architecture (EDMMA) integrated with Apache Kafka, applied to a real-time AI-powered real estate platform. The proposed architecture is rigorously compared against architectural approaches and experimental results reported in recent scientific literature, including Fowler’s monolithic model, Newman’s microservices architecture, Brown’s modular monolith, and multiple event-driven and Kafka-based distributed systems studies. Quantitative results demonstrate that the proposed architecture achieves up to 38% lower response latency, 45–55% faster recovery time, and approximately 60% lower operational complexity compared to microservices-based systems, while maintaining near-equivalent throughput. The findings confirm that event-driven modular monoliths represent a superior and pragmatic architectural choice for real-time, data-intensive domains.