Solving the reproducibility crisis in machine learning

The AI world is facing a dilemma: Machine learning model results are very hard to replicate. I worked with a startup tackling this problem and designed their first prototype: an interface in which engineers can plug in their model, share it, and compare it accurately against others. To take on this project, I dove headfirst into the AI rabbit hole and worked closely with ML engineers to create a viable user experience.

A section of a mockup of a page detailing a machine learning model focused on image recognition A section of a mockup of a page detailing a machine learning model focused on image recognition A section of a mockup of a page detailing a machine learning model focused on image recognition A section of a mockup of a page detailing a machine learning model focused on image recognition
A mockup of a page detailing a machine learning model focused on pitch-tracking
A section of a mockup for a pitch-tracking machine learning model, showing the music player and output graphs
A mockup of the 'Browse models' page for this project, showing rows of different ML models to click on A mockup of an ML model's 'History' tab, showing code commits and the model's improvement over time
A screenshot of UX flow work for this project