TripleBlind: Privacy Suite

Privacy Suite, TripleBlind’s flagship product, had been entirely designed by the engineers coding it prior to my joining the team. As such it lacked some basic best practices and finesse that mature apps require. When users log in, they were thrown into a flow to add a dataset to the product, which wasn’t even the first link in the navigation.

RESPONSIBILITIES — Product Manager, Conducted Stakeholder Interviews, Developed User Stories, Designed User Interface, Approved New Features for Release
1
2
3
1

Use of alert colors in primary navigation. Change color so as not to inadvertently make users think there is something wrong.

2

Illogical starting point for newly logged in users. Create welcome screen with quick links.

3

Mixed alignment on forms. Left align for visual consistency.

Working with partner organizations we determined that users would like to see an overview of progress in a dashboard style interface, this became the natural welcome screen for users.

We give them a brief overview of the number of assets (datasets and algorithms), active processes (calculations or model trainings), bookmarks and collections they have access to.

From their users can see recent activity. That might be datasets they explored or processes they’ve started, as well as a list of their bookmarks.

The feature was well received by partners who appreciated the ability to quickly revisit previous work without having to hunt through menus for it.

User Management

One piece of feedback we received in nearly every POC we ran with our partners revealed frustration with our approach to user management. It was clear to us that the concepts underlying our system had a bit of a learning curve and IT administrators wanted a simple way to onboard and bucket users.

The answer here was clear after researching the approaches taken by modern data catalogues. Data catalogues are a good analog to our product, as we were simply creating an open catalog, or marketplace. Data catalogues leveraged role based access control that mapped to enterprise data science rolls. We adopted a similar approach that allows admins to assign the role of Data Steward, Data Worker, Data Scientist, or Administrator to users.

Not wanting to limit our partners, we also designed a method that would allow administrators to create custom roles with any mix of permissions they wished to include.

AI Process Builder

TripleBlind came with a robust SDK that allowed the most skilled of data scientists and researchers to directly write python code for full access to all our capabilities. The web UI however was severely limited in capabilities by comparison. Advanced techniques like ML training had to be done via the SDK.

It became clear through enterprise POCs that the expectation for this kind of tool was no-code/low-code and when the coolest things your product can do require a PhD to leverage, you need to invest in your experience.

Our lead data scientist and I prototyped a canvas solution that would allow anyone with minimal training to perform any number advanced data-science tasks without writing any code. In fact, we write the code for you, as often this is required during the submission process when getting diagnostic algorithms approved for clinical use.

See More