Why should you care?

AIPredict already runs on ASP.NET Core — ML.NET keeps latency and deploy simple on CPU.

See it live — copy this example

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

// Same process as the API
var engine = ml.Model.CreatePredictionEngine<Tx, FraudScore>(model);
var score = engine.Predict(new Tx { Amount = 1200 });
Console.WriteLine(score.Probability);

What happened?

• Pros: C#, PredictionEngine, ONNX option.
• Cons: not every deep-learning research model — use ONNX/TF when needed.

Practice next

1. List one API that could score in-process.
2. Note when you would call Azure ML instead.
3. Compare to a Python sidecar.
4. Sketch fraud API calling PredictionEngine.
5. Write one con of ML.NET.

Remember

In-process scoring. C# end-to-end. ONNX when needed.

AIPredict in-process scoring

Fraud API scores on App Service CPU.

Outcome: No extra Python hop for v1.