GRAVITY AI
A centralized marketplace to buy and sell enterprise-level algorithms
Project Overview
Timeline: 15 days
​
Team: Suyoung Min, Julia Sevilla
​
Role: Scrum master, UX Researcher, Interviews, Usability Testing
​
Tools: Sketch, InVision, Respondent.io
The process of selling an algorithm is not a simple one. Data scientists have difficulty validating the value proposition for their algorithms.
​
In this project, my team and I tackled the issue of creating a marketplace for data scientists to showcase and sell their algorithms.
Kick-Off Meeting
Before getting into our research, we met with our client, Dan, the founder of Gravity AI to understand the problem and subject matter. Because we were unfamiliar with the space, we asked questions to get a better understanding of the task at hand. We made sure all of our goals were aligned, the timeline we were working with and what is within the scope.
​
Dan had informed us that there were workflow problems that involved uploading algorithms and navigating information. We were also informed that the buyer portal was set up and a seller portal was needed.
Client Testimonial
"You did an incredible job and we will definitely be building from your learning and approach" -Nicole Rufuku, Head of Product
The Challenge
Initial Hypothesis
My team was given the task to design a seller's portal with the primary focus of:
​
-
Sign up / Onboarding
-
User flow for uploading algorithms
-
A community for sellers
"Third party data scientists have a hard time providing enterprise-ready access to their algorithms"
Discover
User Surveys / Interview
Seller Opportunity
6/6
Users found the platform useful, however, were unclear about the logistics
We conducted two rounds of surveys. The first time we targeted people who were had experience with algorithms along with people who have experience selling algorithms. In the second round, we targeted data scientists specifically with familiarity with the subject matter, which was buying and selling algorithms.
​
We received 4 responses from the first round and 5 from the second round and got valuable insights from all of our survey results. We tried to get as much insight and learn as much as we could about the subject matter.
Visuals
Data Sets
6/6
Users needed a visual representation of the data they were looking at so that it would be easily digestible
4/6
Users mentioned that datasets are necessary to test algorithms to ensure the right solutions
Feedback
5/6
Users also wanted a way to get feedback so they can continue to improve their product
"I need to know if my algorithm is running correctly"
Competitive Analysis
Because​ the problem space is relatively new, our competitive analysis analyzed existing and running websites that had an onboarding process, discussion board/forum, and a dashboard. At this point, we needed to know what else was out on the market and tried to learn from them.
​
We researched Algorithmia and SingularityNET as competitors and GitHub as a comparator.
Define
User Persona
The user persona was created from the key findings from our team's research and interviews. Based on the combined findings gathered from our research and interviews with data scientists and subject matter experts, we created our persona to help guide our design decisions.
​
Meet Artemis Petrov! Artemis is a 45 year old data scientist with the goal to sell his algorithms. He also needs feedback from buyers or other data scientists and a nice way of organizing how his algorithms are performing.
User Flow
With the research complete and the problem and solution defined, we proceeded to create the user flow, which served as a skeleton to guide the screens that made up the prototype.
Develop
Problem Statement
How might we help Artemis communicate the value proposition of his AI algorithm on the platform so that enterprises can understand and ultimately make a purchase?
Service Design Blueprint
It was important that we understood the overall workflow before we started designing a solution. We created a service design blueprint to see how the platform would be working on the back end to get what the users needed. The service blueprint helped us communicate how my team would deliver the experience and how Gravity AI would function. This allowed my team to address potential organizational pain points and the breakdown of the internal process
Wireframes
Low-Fidelity Prototype (Paper Sketches)
We sketched our low-fidelity design prototype to determine information architecture, key features, functionalities, and user flows.
As we continued our research we moved to mid-fidelity prototype and went through various design iterations.
Deliver
Final Design
Sign In
This is the first screen that users would see when they sign into their account
Dashboard
Users would see this screen once they sign in. This would show their algorithms and how the algorithm is performing.
Upload Algorithm
This is where the algorithm would be uploaded whether it locally on Git and pushed your algorithm to our link or work through the Web IDE
Record Demo
A major feature was being able to record a demo. Being able to show their demo to stakeholders was effective. Users will directly input their data and record the results
Visuals
A visual representation was a major concern for our users. They wanted to track how much money they were making and interested in what industry was using their algorithm along with the conversion rate.
Usability Testing Report
Onboarding
Score: Minor Issue
Upload
Score: Minor Issue
Easy of Use
Score: Success
-
2 / 3 Users thought the homepage was directed more towards the buyers and not so much the sellers.
-
2 / 3 Users were not clear on setting up the account page for payment details.
-
3 / 3 Users able to complete their tasks.
Next Steps
1. Further testing - Test on potential sellers and reiterate on the design process
​
2. Security measures - Data scientists want to know who is using/where their algorithms are being used
​
3. Rating system - Who should be rating the algorithms?
​
4. Fee negotiations - Work with Gravity AI to determine cost logistics