MACHINE LEARNING DATABASE

Background

The background of this project was very straightforward. Since I worked on it on my summer internship at the company Sonalysts, I had been given the prompt. Although some features were to my discretion, the overall goal of the project was established prior to my arrival. Some features are not depicted in this entry due to company confidentiality.

​

Research

The bulk of this project was user research. I started the project with some very basic research into what the functions of the project had to be. I met with the project lead and a front end developer within my department to get a better idea of what features should be included in the final version. From here, I also researched which type of users would be interested in the service in order to develop personas. I also did some analogous research into similar products to see which features were most important. Using all this information, I made an experience map to visualize which users would go through which processes in which order. This map helped a great deal when considering which screens had to be designed. After this basic research, I made a basic wireframe of the service in Balsamiq.

​

From here, I interviewed 10 people in different departments. I asked them first which type of persona they identify with, and then asked them stakeholder specific questions (for instance, I asked the researchers setting up the experiments different questions from those who would just seek out data). I then showed them the Balsamiq rough draft and got their opinions on that. From there, I gathered many important insights that I presented to the department. After that presentation, I started to make a survey, because I had realized that there were some gaps in my research. This survey was sent out to any interested individuals in the company, and I received approximately 15 responses. At this point, I had also switched to Adobe XD and begun implementing some of the changes that were suggested to me in the interviews. From there, I implemented changes suggested through the surveys as responses arrived.

​

Prototyping

The prototyping of this service was fairly straightforward. As I indicated earlier, the prototype was largely dictated by the experience map that I had made. It contains the four main personas: Users (who will be viewing and downloading data), Contributors (who will be tagging data), Developers (who will be creating and running data experiments), and Admins (who will be overseeing those who are running data experiments).

​

The following is an overview of the mapped out process: users and admins upload raw data to a data lake. Other users can search for, view, and download datasets to support their work or passion projects. Developers/researchers can create experiments for any data set, whereby other users/contributors volunteer to tag data (e.g. drawing bounding boxes or identifying entities in imagery, declaring the sentiment or intent in text, etc.). The tagged data resides in the data lake, where it can be downloaded and used in artificial intelligence development. Admins oversee this process and edit or delete datasets or experiments as needed.

​

Specific changes I had made to my prototype after research include three main elements: Related content (when looking at a data set for project research, users are shown similar datasets), Author information (for Sonalysts partners to see which coworkers have worked with which subjects), and Data context and summary (to know the context why data may have been taken, as well as to have a quick summary of certain pieces of information). After gathering those insights, alongside a handful of smaller takeaways, I improved upon my original design.