You can read more about how it was created here.
Note: This demo loads a trained model into the browser using ONNX Runtime and performs inference directly in the browser.