What is this project about?

This one was about training models through machine learning to predict, classify documents and turn into real data from images. The more samples you give to the model, the more accurate it becomes. But what could I give some structure to something so analytical?

We had to build the graphic panel to teach and train the models, then the models were fully trained become into pipelines could take from minutes to hours to process the samples. Later, the predictions to test the models. It took several meetings and a weekly follow up to get the structure, and patterns for each step.

The UI: relevant information for the user

Structuring the process and organizing the information, incomes and outcomes into steps, helped to also organize the UI. The main challenge was the amount of information, the extremely large labels outcoming from the training. What was from overall the actually relevant information for the customer, and what was the data for the model?

For example, the predictions were secured by encryption, these ones were stored in larges keys (hash), but the user should copy those to use them afterwards. The key itself was meaningless in a semantic way but too large for a list. The solution was to truncate the hash in the UI. Adding a copy function as a trailing icon to copy the content without having to highlight the whole hash with the cursor.

Another example was the SL Annotator was inspired by simple graphics software (like the old friend Paint) familiar for users to draw polygons and catch the zones inside the files to train the models.