The increasing reliance on cloud computing has increased the demand for service stability and dependability, posing problems for Site dependability Engineers (SREs) responsible with proactive monitoring and responding to possible incidents. Despite advances in monitoring tools, SREs struggle to find anomalies in a timely manner, resulting in service failures and a bad impact on customer perception, necessitating the use of an automated anomaly detection system. In this paper, Disentangled Graph Variational Auto-Encoder Based Framework for Intelligent Anomaly Detection to Strengthen Site Reliability Engineering and Improve Operational Efficiency in Cloud Computing Environments is proposed. Large Language Models is used to recognize basic components of cloud infrastructure, their fault modes and behavioural patterns, forming the basis of an innovative anomaly modeling approach. Initially, the data are collected from Infrastructure as a Service multizone regions. The gathered data is saved in Cloud Object Storage. Disentangled Graph Variational Auto-Encoder is utilized to detect anomalies by modeling the intricate relationships and identifying abnormal patterns across the cloud infrastructure environment. Then the proposed is implemented and performance metrics such as Memory Usage, Throughput and Latency are analyzed. Finally, the proposed method provides 26.68%, 25.75%, and 26.16% lower memory usage and attains 29.08%, 30.70%, and 16.26% higher throughput compared with existing methods: Reliability-aware web service composition along cost minimization perspective: a multi-objective particle swarm optimization in multi-cloud scenarios, Strategies for effectual resource management in federated cloud environs supporting Infrastructure as a Service and A Model-Based Schemes Engineering Plugin for Cloud Safety Architecture Design respectively.