Platform Engineer 2026

AI Governance Platform — EKS

FastAPI inference service deployed on Amazon EKS demonstrating production AI governance: per-request audit logging, machine-readable model card with EU AI Act classification and fairness metrics, Helm chart with horizontal pod autoscaling, and Terraform-provisioned EKS cluster with IRSA.

FastAPIEKSHelmTerraformPythonscikit-learnDocker
Source Code

AI governance platform showing how to operationalise responsible AI requirements on Kubernetes — audit trails, model cards, fairness metadata, and regulatory classification baked into the service layer.

What it does

  • /predict — churn inference with per-request audit log entry and unique request_id for compliance traceability
  • /governance/model-card — machine-readable model card with version, metrics, fairness scores, EU AI Act tier, and review date
  • /governance/audit-log — full prediction audit trail queryable by downstream compliance tooling
  • HPA scales 2–10 pods on CPU (70%) and memory (80%) thresholds, both overridable per environment
  • 23 unit tests (TDD) covering inference, audit log, model card, and health probes

Technical highlights

  • POPIA/RICA-aware audit design: features and predictions logged, raw PII excluded
  • Fairness metrics (demographic parity, equalized odds) computed in the SageMaker eval pipeline and published to the model card per training run
  • enable_irsa = true in Terraform — pods assume IAM roles via OIDC, no static credentials
  • Multi-stage Docker build: 146 MB image, non-root user, baked-in health check
  • Helm _helpers.tpl with fullname, labels, selectorLabels — lint-clean chart