Why should you care?

AIPredict retrains fraud nightly and bulk-scores transactions without blocking checkout APIs.

See it live — copy this example

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

public class FraudRetrainWorker : BackgroundService
{
    protected override async Task ExecuteAsync(CancellationToken stoppingToken)
    {
        while (!stoppingToken.IsCancellationRequested)
        {
            var model = pipeline.Fit(await LoadTrainSetAsync());
            ml.Model.Save(model, schema, "models/fraud-nightly.zip");
            await Task.Delay(TimeSpan.FromHours(24), stoppingToken);
        }
    }
}
builder.Services.AddHostedService<FraudRetrainWorker>();

What happened?

• HostedService loops on schedule: load data, Fit, Save zip.
• APIs pick up new file via watchForChanges or deploy hook.

Practice next

1. build BackgroundService.
2. Fit and Save on schedule.
3. Register AddHostedService.
4. Add cancellation on shutdown.
5. Upload zip to blob after Save.

Remember

Train off request thread. Scheduled retrain. Save new zip.

AIPredict nightly retrain

Worker writes fraud-nightly.zip at 2am.

Outcome: API reloads without manual deploy.