Multi-Stage Research to Build a Machine Learning Platform
My team and I partnered with Aspinity, a machine learning start-up, for our capstone project for the Master of Human-Computer Interaction program at Carnegie Mellon University.
We began by researching the problem space of analogML thoroughly and defining key opportunity gaps: education, collaboration, and workflow efficiency.
We then ideated and tested potential solutions that led to the creation of an interactive prototype with service and design recommendations.
I co-led the UX research planning process for this project. In addition, I conducted and analyzed the research sessions, alongside other team members.
I was one of two UX research leads for the project. The rest of the team comprised of 1 project manager and 2 product designers.
Aspinity has developed an innovative technology, analogML, which notably contrasts with traditional, digital machine learning. Few people know how to create analogML applications besides Aspinity's internal engineers. Aspinity wants to empower their client's engineers to develop their own analogML applications but weren't sure how to prepare them for the complex task.
Our team kicked off the research process by conducting secondary research to understand the complex space we'd be working in and inform our initial approach.
We conducted interviews and contextual inquiries to grasp the mental models, needs, and pain points, of ML engineers and data scientists who may create analogML applications.
By using prior research, the team generated multiple initial concepts that were evaluated through speed dating sessions.
In parallel to these sessions, we had users think-aloud while completing tasks on a competitor's platform.
With the insights delivered by these methods, we understood how different concepts addressed users' needs. This knowledge enabled us to refine our approach and identify additional opportunities to consider.
Once the team identified a solution that resonated with users, we designed the initial iteration. The solution was evaluated through a series of usability tests to identify areas for improvement and inform the design of future iterations.
The research insights guided our ideation and iteration process, resulting in the creation of a platform that offers a comprehensive ecosystem of support.
This ecosystem empowers client engineers by facilitating self-service, fostering a community for learning and engagement, and providing access to expert assistance, thereby enhancing their overall experience and effectiveness.
When we presented the final solution to our client, we were thrilled to witness their enthusiasm and satisfaction with our approach. The client's positive reception indicated a strong alignment between our solution and their needs.
As a result, we are optimistic that the project is likely to move forward into development, marking a promising step towards realizing our vision of empowering client engineers to create their own analogML applications.
If I were to pinpoint an area for improvement, it would be the technical evaluation of our platform with users.
Since prototyping tools have technical limitations, our design was incapable of running actual code, which is an integral element of the platform we designed.
To address this, we resorted to Wizard of Oz prototype testing as a means to simulate the platform's functionality and gauge user reactions.
Unfortunately, this approach proved challenging to execute, leading to a limited number of sessions.
If our team had more time and resources, I would have liked to
find a better way for us as researchers to simulate the technical processes, such as running code on the platform.
Such a step would have further enriched the depth and accuracy of our research findings.
