Miguel Malfavon

Helping Google Research collect data at scale: Designing an app from 0-1

 

*Mocks are representational of final design

 

TL;DR

  • I worked under tech, time, and budget constraints to take an app concept from 0-1 in three months.

  • I did desk research, created concept sketches, interviewed stakeholders, developed wireframes, and designed high fidelity mocks.

  • I partnered with a UXE to consult in developing a working prototype.

My team: AIUX, Google Research

My role: UX Designer

Partner team: AI Test Kitchen, Labs

Duration: 3 months

Background

Sundar unveils AI Test Kitchen at Google I/O 2022

AI Test Kitchen is an app that offers the public fun, quirky demos powered by Google's latest language and generative models.

I identified the opportunity to reframe the app’s value proposition from pure delight, to a platform that’s delightful and also helps Google Research test hypothesis at scale.

Challenge

In 3 months, identify the highest value concept to design, and prototype a bite-sized demo that delights users and tests hypothesis from Google Research.

Process

 

  • To better understand the ingredients that made a successful demo, I co-led a brainstorming session around our team's assumptions for what a good demo concept should be.

    After I had a sense of what a good dish should be I organized another brainstorming session, this time with the larger AIUX design group, to generate more concepts.

    I primed the group with context and constraints around what the dishes could and couldn't be and started a Crazy 8's brainstorm.

  • I then used the brainstorming ideas as inspiration to sketch additional refined concepts into short storyboards.

    Next, I placed my sketches into a presentation deck and added further context to help explain the unique magic each concept gives the user.

    Through a dot voting exercise the team selected Board of Advisors as the concept to move forward with, as it showed the most promise to gather meaningful data from.

  • After completing the final mocks I switched focus to helping our team's engineer create the dog personalities in MakerSuite.

    In short, you need to feed the model some example data to train its responses on. So, I made a spreadsheet where I wrote short stories and responses from the perspective of each dog in the tone of voice their feedback and suggestions should appear, and we input that as the training data.

 

Solution

 
 

*Mocks are representational of final design

 

Designed and socialized the concept across Google Research.

The demo is about a group of AI agents with different personalities and perspectives that help the user ideate, riff, and refine their work.

Throughout the experience, the user collaborates with the agents in writing a short story, gaining new ideas and perspectives from them along the way.