This conference talk explores how to implement anomaly detection for observability from day one of application deployment. Learn how to ensure seamless performance monitoring without waiting for historical data to train ML models. Prashant Gupta and Kruthika Prasanna Simha from Apple demonstrate how to leverage pre-trained ML models for immediate anomaly detection when launching applications on cloud platforms. Discover techniques for deploying lightweight, unsupervised models using cloud-native tools like Kubeflow, enabling real-time identification of potential issues before they escalate. The presentation covers practical methods for setting up and fine-tuning these models to effectively monitor application health metrics from the very first deployment, enhancing system reliability from the start.