MITRE Corp x Cornell Tech:
ARGO Data Marketplace
Allowing for secure confidential data sharing and commercialization amongst institutions.
OVERVIEW
Partnered with MITRE Corporation, we built Argo, a B2B data commercialization marketplace that supports confidential data sharing using blind learning technology from TripleBlind®. This allows organizations (such as government, academic institutions, private/commercial firms, and public sectors) to commercialize on confidential data that was previously not accessible.
COLLABORATORS
1 PM, 1 DS, 3 SDEs, MITRE Corp advisors
WORK
SaaS Web Design, Visual Design, Research
TOOLS
Figma, Illustrator, Photoshop, Miro, Trello
DURATION
February 2022 (design) - Jan 2023 (launch)
Vision
Designing and building Argo, a web data commercialization marketplace that supports confidential data sharing using blind learning technology. The primary users are organizations such as government, academic institutions, private or commercial firms, and public sectors.
My responsibilities
Our team consists of a PM, a designer, a data scientist and four engineers. I was primarily responsible for crafting the design system, flows and hi-fi prototypes. In the meantime, we all worked together from end to end, including on researching, ideating, designing and ultimately shipping. At the end of our collaboration, I delivered design demos and my team pre-launched applications for MITRE who was able to conduct internal testing with its verified users.
Goal
Designing a data commercialization marketplace that displays the meta-information of confidential data without violating privacy issues.
User groups
data provider
register and share dataset; can also be data user
data user
request and access dataset application
Dataset types
-
Registered datasets are provided by data provider and available on marketplace
-
Accessible datasets are for data user who can request to access their info;
-
Available datasets are for data user whose requested datasets have been trained by blind learning, so they can access then.
Outcome
Introducing Argo:
Our solution aims to support and optimize confidential data sharing amongst organizations with a web-based marketplace aptly named Argo using blind learning technology, in the meantime maintaining data confidentiality. Argo's value proposition outlines in the three design spaces:
Research
As the entry point to explore the intersection between enterprise design and data domain, my team and I created a research plan to document our initial goals from different angles: primary research interviews (experts and stakeholders), secondary literature reviews. They are key to help us review pain points from customers regarding confidential data processing because there were no prior products dedicated to this challenge.
Understand the background
The existing federated and split learning face two main challenges:
-
it is still profoundly challenging to share and utilize such sensitive data due to user privacy, ethical and legal concerns;
-
existing methods could lead to impractical training time.
Source: TripleBlind, Inc. (who provides blind learning technology to empower Argo marketplace)
Insights - decompose
We wanted feedback from MITRE on which stakeholders were likely to be involved in sharing and selling data. We also learned more about blind learning tools developed by MITRE's partner, TripleBlind in which this can make it possible to train new models on remote data and run inference on existing models, while protecting the privacy and fidelity of data and intellectual property.
Data access
Users can preview existing datasets, register new ones from local, and access each categorized dataset by requesting with MITRE credits
Marketplace
A web commercialization platform that is beneficial for organizations to trade confidential data and algorithms
Privacy
Users can trust Argo on data trading while maintaining data confidentiality that would otherwise prevent sharing
Blind learning
An AI tool that can train new models on remote data and run inference on existing models while protecting data privacy and fidelity
Synthesis
Mapping
Our team worked with MITRE advisors to synthesize insights from interviews by mapping these notes we took into feature categories. We used the tool Miro in this stage, and due to some NDA I showed the results of our process on what we would like to implement in the application.
Browse the recommended datasets on Argo platform
Define detailed access (organization types and property preferences)
Rate and review datasets to provide a reliable reference
Register datasets to Argo marketplace for sharing and commercializting
View analytics of each dataset such as included files and feature illustrations
Add datasets into shopping cart and check out to complete the request
Experience flow
After successfully curating our features and meeting with our advisors, we organized them into a structured user flow that outlined Argo marketplace. We projected what Argo would look like after introducing new features to make sure the model could scale to our product vision.
Vision concept + RITE
At early stage, I crafted two screens both showing the analytics and details of a dataset, and ran a few RITEs (Rapid Iterative Testing and Evaluation) to effectively discover issues and identified the better option from users. We presented to our MITRE advisors who ultimately voted the second option because it can quickly filter the selection and allows for viewing different datasets in an interactive menu.
Mid-fi iterations allow to quickly identify the strength for each screen by focusing on layouts
Visual guildeline
Argo marketplace processes its algorithms from MITRE, deep learning techniques from MITRE's partner TripleBlind, so its visual system follows both partners' brand to build a data-blue-driven, consistent experience for users.
End-to-end
solution
Our marketplace prototype consisted of web interface that included screens like signing in, homepage, dashboard, register dataset, dataset rating and reviews, and finally checkout.
Key persona
We created a representative persona modeled after interviews with users previously.
Argo marketplace's secure data sharing experience helps Eric accomplish his goal of commercializing and accessing sources of needed datasets by offering a secure sign-in process, dataset registering with only retaining meta-information, as well as allowing for secure commercializing while having transparent but reliable rating system that validates the usability for each dataset.
Sign in with SSO
Eric is a verified user of MITRE and he could sign into Argo with SSO from his organization (Cornell). in order to ensure his sign-in security.
Register a dataset
Eric wants to register a Cornell-owned dataset on Argo. There are two steps needed: (1) select and upload own dataset, and (2) define its access. Argo will make sure to only collect meta information of his dataset during the registration.
View dataset details
Eric can review the detailed information of his registered dataset empowered by blind learning technology. In the meanwhile, he is opt to editing the contexts if needed (by selecting the edit button), while easily browsing the in-app menu to find out other interesting datasets.
Dashboard
Eric can access comprehensive information from the dashboard named under "My Datasets" such as the smart analysis of datasets (in the unit of his organization) in profile, access requests and register history.
Rate a dataset
Eric is able to rate and review a dataset after requesting it, and provide real and transparent feedback that can help others evaluate its performance, usability.
Check out
Back to homepage, Eric can add any available datasets into shopping cart, where he will review and add additional datasets before making the final payment on the external page operated by MITRE.
Task analysis
We wanted to validate the design and under our customer's workflows. We moved forward to the last phase: task analysis for Argo marketplace. We spoke again with our MITRE advisors to go through a clickable prototype with 5 tasks and report if they think that 1) they completed every task, and 2) their thoughts and feedback. We were able to identify areas that were successful and those that required further iterations in the next steps.
Reflection
Next steps
How can we design to enable training on multiple confidential datasets from different sources while still maintaining their data security?
Takeaways
What makes Argo a special project is its system-driven, complexity and large scope. I learned to build a mental model for large user groups, other than individual consumers, who can share and commercialize confidential data on a secured marketplace backed by blind learning technology.
-
🚀clear mentality on complex design challenges
-
🧠open to share incomplete work & proactively ask for feedback