Agentic AI for Microservices: Autonomous Optimization of High-Volume Financial Transactions in Cloud Native Environments

This paper presents a Deep Q-Network (DQN)-based Agentic AI framework for optimizing performance in cloud-native microservices, to handle high-volume financial transactions. The system comprises of independent agents for three crucial tasks: load balancing, predictive caching, and auto-scaling. A synthetic FinTech transaction dataset was generated to simulate real-world trading patterns, traffic bursts, and workloads. The proposed agents were trained and evaluated against traditional RoundRobin, which is a rule-based baseline under identical conditions. Test results show that this DQN-based approach reduces average latency by 47 %, increases cache hit rate by 34 %, and cuts scaling operations by 48 %. These improvements indicated that Agentic AI could help financial systems to be more reliable. The study also discusses implementation details, training challenges, and future extensions involving multi-agent learning and real-world deployments.