Hierarchical Expert Multi-Agent Framework for Causal Root Cause Localization in Cloud-Native Microservices

Cloud-native microservices have high complexity because they have dynamic dependencies, heterogeneous monitoring data, and many types of failures. This makes root cause localization a challenge. Existing methods often cannot balance accuracy and low latency, and they struggle with new failures, large system size, and multimodal data. This paper presents HEMA-RCL, a hierarchical expert multi-agent framework that uses different large models for collaborative diagnosis under complex dependencies. The framework uses layered expert agents led by a global orchestrator. It adds dynamic agent generation through efficient low-rank adaptation. It reaches agreement with belief propagation and causal enhancement to reduce hub misidentification. It also unifies multimodal data through temporal alignment and robust feature engineering. It applies context-aware prompt optimization to reduce hallucination in large models. HEMA-RCL improves on prior methods by enabling accurate, scalable, and efficient root cause localization in cloud-native microservice systems.