Follow the workload, not the tool list
Infrastructure becomes easier when each component answers one operational question.
Swipe or scroll sideways to read the full diagram.
Source code builds an immutable image stored in a registry. Kubernetes schedules Pods on nodes whose runtimes pull the image; Services send requests to ready Pods. Workload autoscalers change replica demand, node autoscalers supply machines, and telemetry informs those decisions.
The pages here own the reusable mechanics. Artificial Intelligence (AI) Engineering links back when a large language model (LLM) changes the resource profile, startup path, or scaling signal.
Start with the layers
- ContainersImages, registries, runtimes, and graphics processing unit (GPU) access.
- KubernetesPods, nodes, Deployments, Services, storage, init containers, and probes.
- ScalingWorkload replicas versus machine capacity, with the Horizontal Pod Autoscaler (HPA), Kubernetes Event-driven Autoscaling (KEDA), and node autoscaling.
- Operations & observabilityMetrics, logs, traces, health, and the evidence used during incidents.
- Cloud architecture, IAM & costService identities, secure request paths, managed data services and cost trade-offs.