An Agent-Orchestrated Anomaly Detection Pipeline for Network Performance

Abstract

Today's telecom anomaly detection relies on hand-engineered pipelines that break whenever data or scenario shifts. We propose a reasoning agent that builds, tunes, and explains its own detection pipeline by orchestrating tools in an adaptive loop, thereby turning a brittle process into an auditable one and laying the foundation for agentic reasoning on cellular networks.