An interface for supervising Gamalon's machine learning in real time.


contextual inquiry -> persona creation -> low-fidelity prototyping -> service blueprinting -> parallel prototyping -> card sorting -> high-fidelity prototyping -> production


Jon Linton, Alan deLespinasse, Ira Ladson, Ben Vigoda, Pawel Zimoch


2019 / 4 months

Problem Statement

Machine learning can seem like a black box. This is because it is often impossible to inspect and edit algorithms. As a consequence, algorithmic bias is difficult to identify and root out.


Gamalon Thesaurus is a web-based application that allows data scientists at Gamalon to supervise machine learning processes in real time. Using Thesaurus, a data scientist can inspect hierarchical natural language models as they are being assembled. They can also step in to make adjustments and correct mistakes.

Selected Assets

A selection of assets from the prototyping process are presented here. For a detailed case study, see this page.

[Figure 1] A service blueprint mapping out Gamalon's plan for human-machine collaboration.

[Figure 2] An early prototype showing a user interacting with a single piece of natural-language data, i.e. a customer message.

[Figure 3] An early prototype showing the model assembly process via clustered groups of data that occur together.

[Figure 4] A later prototype stripped of natural-language datapoints, showing only a branch of a hierarchical model (left panel) and its constituent ideas (right panel).

[Figure 5] A later interface showing the hierarchical model (left panel), two constituent branches of the model (center panel), and constituent ideas (right panel).

[Figure 6] A speeded-up video showing a data scientist collaborating with Thesaurus to build a language model about edible consumer products. Skip ahead and pause at 3:19 to see the results of the model-building exercise.