Azure AKS ML Prediction Service demonstrates a production-grade Kubernetes architecture for serving LightGBM models via FastAPI pods behind an NGINX ingress controller and Azure Load Balancer. User traffic flows through the public load balancer to the ingress controller, which routes requests to ClusterIP services distributing load across multiple FastAPI replicas that invoke the shared LightGBM model. Prometheus monitors pod metrics and health endpoints while Grafana visualizes performance, enabling observability across the inference pipeline. Infrastructure is provisioned via Terraform with separate staging and production AKS workspaces, container images built by Azure DevOps pipelines and stored in Azure Container Registry, and state managed securely in Azure Blob Storage. This architecture exemplifies best practices for high-availability ML serving: multi-pod redundancy, managed identity-based ACR authentication, infrastructure-as-code deployment, and integrated monitoring. Fork this diagram on Diagrams.so to customize namespaces, scaling policies, or add additional monitoring components for your ML workloads.