Cloud-native microservices architectures have become essential for developing scalable, flexible, and resilient distributed applications. However, their dynamic nature, complex service dependencies, and distributed operational environments introduce significant challenges in fault detection, root cause identification, and automated recovery. Conventional monitoring and rule-based fault management approaches often struggle to identify complex anomalies and coordinate effective recovery actions across interconnected services. This study develops an AI-assisted fault detection and self-healing architecture that integrates machine learning, real-time monitoring, intelligent fault diagnosis, and automated service recovery within Kubernetes-based cloud-native environments. The architecture employs Random Forest, Isolation Forest, and XGBoost algorithms to analyze operational metrics, identify abnormal service behavior, and classify faults associated with CPU overload, memory exhaustion, network latency, and service unavailability. A centralized decision-making engine integrates fault classification with automated recovery mechanisms, including container restarting, service scaling, and workload rescheduling. The research methodology incorporates Prometheus-based monitoring, distributed tracing, controlled fault injection, and reproducible experimental evaluation to assess detection accuracy, precision, recall, F1-score, mean time to recovery, and service availability. Comparative evaluation against conventional Kubernetes recovery mechanisms is designed to examine the effectiveness of AI-assisted fault management under different failure conditions. The developed architecture establishes a systematic approach to integrating intelligent fault diagnosis with autonomous recovery workflows, providing a foundation for improving operational resilience, reducing manual intervention, and supporting reliable cloud-native microservices management.