Info
      Info

About this project

What does a 3D machine learning model look like?



In Spring 2023, I teamed up with creative developer Alexander Taylor to design and build the Playground: an interactive demo that showcases a computer vision model called Vauxel.

Vauxel, developed by our colleagues at Loci, translates 3D assets into parsable text such as tags, categories, colors, and more. 

It is the best in the world at this specific job.

The jobs matters because modern life quietly depends on 3D. Most objects made by humans—smartphones, cars, skyscrapers—start out as 3D designs. Today’s biggest movies and video games star 3D characters and environments.  And the technology of tomorrow—from AR to self-driving cars—relies on digital 3D mapping. 

The world runs on billions of 3D assets. Vauxel labels make them easy to search, manage, and use for architects and designers and engineers.

In practice, Vauxel outputs raw JSON. 





Raw JSON output from Vauxel. Awesome, but obscure to most.


Impressive, but obscure. The AI’s wonder is hidden by a language barrier—most people don’t speak JSON. 

Our goal with the Playground was to make Vauxel easy to understand and impressive to look at for anyone. The Playground also needed to let users see how Vauxel performed on different kinds of 3D assets, including ones they uploaded.

Designing around frontier AI like this is exciting because there are no obvious patterns to reference. We kept things simple by focusing on nailing two components: inputting a 3D asset, and viewing the AI output. For the input, we displayed a single, large, spinning 3D asset to highlight the beauty of the medium. Early ideas like displaying multiple assets were scaled back to keep the interface snappy and performant.

Initially, we emphasized user-uploaded assets but realized many visitors didn’t have 3D files handy. To solve this, we included a selection of sample assets and kept the upload option secondary.

For the AI output, we formatted the JSON into clean, readable text while offering a toggle to view the raw code. This balance made Vauxel’s capabilities approachable without losing technical depth. 

 




Design Om Gokhale
Eugene Yi

DevelopmentAlexander Taylor
Jack Davis
Client
Loci
Links
Playground (Beta)

 Cambridge MA
© 2025