AI governance is increasingly a business requirement, not an afterthought. Whether it's the EU AI Act, NIST AI RMF, or an internal risk committee, the question is the same: can you prove your model is behaving as intended, on every request, with a documented audit trail? This post walks through an AI governance platform I built on AWS EKS: a FastAPI service that runs churn inference, records every prediction to an audit log, exposes a machine-readable model card, and packages everything into a Helm chart with horizontal pod autoscaling. Source code : github.com/tsekatm/eks-ai-governance Architecture Client └── POST /predict → FastAPI (EKS pod) │ ├── LogisticRegression inference ├── Audit log entry (request_id, features, result) └── Response: churn_probability, prediction, request_id Kubernetes (EKS) ├── Helm chart → Deployment + Service + HPA (2–10 replicas) └── Terraform → VPC, EKS cluster, node groups, IAM OIDC Governance Layer ├── GET /governance/model-card → version, metrics, fairness, EU AI Act tier └──…