A machine learning framework that speaks C#

What it does

ML.NET is Microsoft’s open-source machine learning framework for the .NET ecosystem. It handles the full model lifecycle—loading data from files or databases, transforming it, training with built-in algorithms, and running inference—without asking developers to context-switch into Python or R. It also acts as a host for existing TensorFlow and ONNX models, so you are not locked into its own trainers.

The interesting bit

The platform support matrix is broader than you might expect: Windows, Linux, and macOS are table stakes, but ARM64, Apple Silicon, and Blazor WebAssembly are also on the list. Running inference in the browser via WASM is a neat trick, though the README warns that some platforms carry specific limitations.

Key highlights

• Native .NET API for classification, forecasting, anomaly detection, and other scenarios
• Can consume TensorFlow and ONNX models for scenarios outside the built-in trainers
• 64-bit support across all platforms; 32-bit Windows works but drops TensorFlow and LightGBM
• Shipped on the same annual cadence as .NET, with servicing branches for bug fixes
• MIT licensed and free for commercial use

Caveats

• Blazor WebAssembly, ARM64, and Apple M1 support come with documented limitations
• 32-bit Windows support excludes TensorFlow and LightGBM functionality

Verdict

Worth evaluating if you are a .NET shop that needs custom models but would rather not maintain a Python sidecar. If you are already productive in PyTorch or scikit-learn, this probably will not pull you away.