International Journal of Science, Engineering and TechnologyAuthors: Sudhir Vishnubhatla
Abstract: Real-time decision making is increasingly vital in domains such as financial services, autonomous systems, healthcare, and large-scale IoT operations. These environments generate massive, continuous data streams where milliseconds can separate success from failure. Systems in these sectors must interpret dynamic information, predict outcomes, and execute appropriate actions without delay. Traditional analytics pipelines, designed primarily for batch processing, lack the responsiveness and adaptability required to function effectively under such conditions. They process data retrospectively, which limits their usefulness when rapid feedback loops or immediate situational awareness are essential. This article introduces a hybrid decision-making framework that integrates Complex Event Processing (CEP), Artificial Intelligence (AI), and human-in-the-loop mechanisms to create adaptive, explainable, and scalable workflows. CEP enables real-time monitoring and detection of significant patterns or anomalies within high-velocity data streams. AI components provide predictive and prescriptive intelligence by learning from evolving data, while the human-in-the-loop aspect ensures oversight, correction, and ethical accountability in uncertain or critical situations. Together, these elements form a resilient, continuously improving architecture capable of sustaining operational integrity under uncertainty. We formalize workflow patterns such as cascade, fallback, and parallel hybrid decision models, each designed to balance the competing demands of latency, accuracy, and interpretability. The cascade model allows lightweight and deep models to cooperate efficiently; fallback logic ensures safety and reliability during uncertain predictions; and parallel decision workflows enhance confidence through consensus across multiple inference engines. The framework’s flexibility allows it to be tailored to diverse operational settings where data context, urgency, and decision risk vary significantly. The proposed architecture is illustrated through a real-time fraud detection use case, highlighting its ability to combine streaming event analysis, adaptive AI scoring, and human judgment. This example demonstrates how the system can maintain a balance between speed and quality of decisions while adapting to changing conditions, ultimately achieving greater robustness, transparency, and trust in high-stakes environments.
DOI: http://doi.org/10.5281/zenodo.17471901
