ODNtuk
← All work Meta AI · Project

Turning AI research into demos people could try

As design manager, I oversaw the team that turned research models into AI-powered demos. I steered the projects by providing guidance, suggestions and feedback, and helped those demos find partner teams inside Meta.

Role
Product Design Manager
Timeline
2020 to 2022
Team
Up to 7 direct reports
Partners
Instagram, Hugging Face, research and engineering
Walkthrough of the PyTorch Live app on a laptop

Strategy

Building AI-powered demos around research models

Research teams had promising models and no easy way to show what they could do. We set up a simple path from idea to shipped demo.

  1. PitchTeams or individuals pair an idea with assets that sell the story.
  2. VetIdeas are weighed on effort, budget and other constraints.
  3. BuildFull demos built cross-functionally with research and engineering, then released.
0 to 1 demos delivered in three months
2
demos launched externally
2
budget managed for AI initiatives
$300K

Funded teams like Instagram carried the experiences forward, with vendor partners like Hugging Face.

Research-led demo

Animated Children's Drawing

A research model that brings a child's drawing to life, turned into an experience anyone could try. Upload a drawing, confirm the figure and watch it walk, dance or jump.

The demo went from research model to public launch, and was shared by Meta's CEO in December 2021.

Mark Zuckerberg's social post sharing the Animated Drawings demo, with an animated drawing playing
The CEO's post about the demo, December 2021.

Engineering-led demo

PyTorch Live opened the door to external testing

Internal teams wanted proof before they'd invest in a demo. An external app gave us a way to get it.

  1. Legal approval to test externallyShipping PyTorch Live as a public app let us work with Legal on approval to test our demos outside Meta.
  2. External results as internal validationReal usage outside the company became the evidence inside it.
  3. Partner teams foundThat evidence helped us find teams willing to experiment with the demos.

Design system

Redesigned the internal AI demo studio

AI Studio was the app teams used to show AI demos. The team created a new React-based design system to reskin and enhance it in a six-week redesign.

  • Kicked off an internal design language
  • Updated guidelines for grid, spacing, type and color
The redesigned AI Studio interface

Marketing site

A home that championed our work

Great demos weren't enough if no one could find them. I advocated for an internal marketing site so the team could show what it had built and win partner teams to carry it forward.

  • Featured Big Bet projects in one place
  • Made it easier to bring on partners
  • Raised the visibility of AI demos across Meta
Walkthrough of the internal AI demos marketing site on a laptop
The AIX site, walking through featured demos.