MNIST Digit Classifier API

This is a small app for handwritten digit recognition using a convolutional neural network.

Instructions: Draw a digit in the canvas below and click "Predict!" to see the model's top three predictions, click "Clear" to start over. If the model is inactive, it may take a few seconds to load the predictions. The canvas works better on computers, I'm still working on improving the mobile experience!


Loading...


How does it work? This model was trained on the MNIST dataset in TensorFlow using the Keras API. The model is stored in a MLFlow Server Model Registry, by clicking "Predict", the canvas image is sent to a FastAPI endpoint deployed on Google Cloud Run. This API fetches the model from the MLFlow Model Registry, processes the image, and returns the top three predictions with their probabilities.

MNIST API architecture