Scalable Multi-Stage AI-Integrated Microservices Architecture for Real-Time Stock Anomaly Detection and Alert Generation

Abstract

This work presents a scalable multi-stage architecture designed for real-time detec- tion of abnormal stock market behavior using distributed microservices and integrated AI components. The system combines rule-based evaluation, streaming technical indi- cators, and lightweight machine-learning inference to process continuous market data with low latency. Each stage of the pipeline operates as an independent service, enabling fault isolation, flexible scaling, and efficient message routing. Experimental assessment using replayed financial data streams indicates stable latency characteristics, consistent anomaly-score separation, and reliable throughput under varying loads. The results suggest that the architecture is suitable for deployment in environments that require continuous monitoring, interpretable analytical signals, and timely alert generation for financial decision-support systems.