Cloud infrastructure today supports large-scale digital services, handling tasks such as storage, computation, and traffic routing. These systems must work continuously without interruption. However, unexpected failures, abnormal usage patterns, or system overloads can affect reliability and slow down performance. Traditional monitoring tools often detect problems too late, offer limited automation, and require manual attention to fix faults. To address these issues, this work introduces a real-time monitoring framework designed to observe, detect, and correct problems in cloud infrastructure as they happen. The system uses AI-based anomaly detection models trained on live telemetry data such as CPU load, memory usage, disk performance, and network throughput. Once abnormal behavior is identified, the self-healing module takes action automatically, such as restarting faulty services, re-assigning loads, or isolating affected nodes. The design follows a lightweight architecture to support both large-scale cloud systems and edge clusters. A dashboard gives operators a full view of current performance and health metrics, while the backend engine works in the background to avoid downtime and reduce human involvement. Experimental tests were conducted using real-world cloud workloads and stress simulation environments. Evaluation metrics included anomaly detection accuracy, mean time to recovery (MTTR), false positive rate (FPR), F1-score, Precision and Recall. Results showed up to 94.7% detection accuracy, and faster recovery under heavy load. This solution improves the responsiveness and reliability of cloud services while reducing operational costs and maintenance delays.