Why should you care?

AIPredict fraud service scales pods horizontally during peak without manual VM sizing.

See it live — copy this example

Use a .NET console or Web API project with Microsoft.ML. Run dotnet run after pasting.

// deployment env from ConfigMap
var modelPath = Environment.GetEnvironmentVariable("FRAUD_MODEL_PATH")
    ?? throw new InvalidOperationException("FRAUD_MODEL_PATH required");
services.AddPredictionEnginePool<TxRow, FraudPred>().FromFile(modelPath, watchForChanges: true);
// kubectl: env FRAUD_MODEL_PATH from configMapKey fraud-model-path

What happened?

• ConfigMap holds model path or blob URI; HPA scales on CPU or custom predict latency metric.
• Follow the steps below — typing the code yourself is the fastest way to learn.

Practice next

1. ConfigMap for model path.
2. Register pool FromFile env.
3. HPA on CPU or RPS.
4. Init container download zip from blob.
5. Readiness probe on /health.

Remember

ConfigMap env. Stateless pods. HPA scale.

AIPredict K8s fraud

HPA adds pods on Cyber Monday.

Outcome: Latency stable at 3x traffic.