User friendly machine learning


For Dialogic, I developed a user friendly environment for machine learning. The tools, which are available online through, allow users to quickly set up image classifiers, text classifiers as well as generic machine learning models using custom features and outputs.

Supported features:

  • Image classification: transfer learn image classifiers using custom categories. Uses transfer learning (with customizable model size and parameters) based on MobileNet.
  • Text classification: user definable categories, character based embedding that is resilient to spelling errors.
  • Generic models: user definable features for input and output, including text (word or character based embeddings), numeric values (normalized), booleans and images (using MobileNet embeddings).

The software runs in the browser and uses Tensorflow.js for training and inference on the GPU (if available) through WebGL. The user interface is powered by Vue.

Visualisation of generic classifier with two numeric inputs (shown here as X and Y axes) and three output classes (shown here as colors). The dots represent training cases.

Read more here (in Dutch)