Modern IT infrastructures are distributed and complex, making fault detection and diagnosis more challenging. Traditional monitoring systems rely mainly on dashboards, static thresholds, and manual analysis. These approaches can generate numerous isolated alerts and provide limited support for identifying the root cause of a problem.
This thesis presents the design and implementation of an intelligent monitoring system that combines telemetry collection, anomaly detection, multi-agent coordination, and large language model (LLM)-based diagnostic reasoning. System metrics and application logs are collected and stored in a centralised monitoring infrastructure. Anomaly detection techniques are applied to system metrics to identify unusual behaviour, while logs provide additional diagnostic evidence. When an anomaly is detected, specialised agents cooperate to investigate the incident, gather relevant information, identify potential root causes, and provide the technical operator with diagnostic insights and remediation recommendations.
The diagnostic process is supported by locally deployed LLMs. A retrieval-augmented generation approach provides access to infrastructure-specific knowledge, including technical documentation, expected system behaviour, known failure modes, and operational knowledge. Local deployment reduces the need to transmit potentially sensitive monitoring data to external services.
The architecture was implemented as an end-to-end prototype and evaluated in a controlled, microservice-based environment using injected fault scenarios. Experimental results show that the system can autonomously integrate telemetry collection, anomaly detection, evidence gathering, multi-agent coordination, and diagnostic reasoning into a single workflow. This work demonstrates the feasibility of combining intelligent monitoring with agentic AI to support faster and more structured fault investigations while keeping human operators involved in critical decisions.
parole chiave
Agentic AI, Artificial Intelligence, Distributed Systems, Intelligent monitoring, anomaly detection, multi-agent systems, large language models, retrieval-augmented generation, telemetry, fault diagnosis, root cause analysis
