Google: Helping Data Scientists Organize their Files

Role: Lead Designer & Researcher

Team: Product Manager, 2 Engineers

Launched February 2024

Final Designs

The new page I designed provides a file management system to help data scientists organize, search, and declutter their workspace through three main tools:

What I did

I designed a centralized file management system (like a Google Drive) to help data scientists organize their work. Along the way, I:

Owned the end-to-end design process • Partnered w/ Product, Research, Engineering, & DevRel • Tested designs w/ users • Iterated off feedback • Advocated for user needs • Adapted to new technical constraints • Supported engineers in implementation

1) A centralized search & filter system for locating files

“Bulk delete will save me so much time.”

“Less work for me.”

“Finally! I really needed folders.”

Discovery

Using insights from past user research, I created 3 personas. I worked with the PM to prioritize user needs.

Early Collaboration

To understand the problem space, I partnered with the following roles:

Senior Designer to review the history of this project
Research Lead to review existing research + plan additional research
Product Manager to discuss business goals and business constraints
Developer Relations Engineer to understand the technical context

Defining User Needs

The Product

Kaggle is a Google-owned machine learning community with over 10 million users. Data scientists create and publish their work to Kaggle’s vast library of user-generated content.

Solution & User Feedback

Design Phase 1:

Early Exploration

Initial Ideation

User Flow Mapping

Explored multiple ways to execute each CUJ

We wanted user feedback to inform my design direction going forward. I prepared four paper prototypes with varying features and information architecture to test on users.

Option A introduced folders
Option B organized content by type
Option C incorporated analytics
Option D auto-organized content by project

Applying feedback & solving design challenges

Feedback from concept testing provided me clear direction on which features users needed the most. Now, it was time to share my insights with the project team, align on new design priorities, and dive into solving complex UI and interaction challenges.

The Problem

Creators on Kaggle have so many loose files that they can’t locate or manage their work efficiently. This slows down their workflow and makes creating content on Kaggle harder.

“Can’t manage my files effectively” has been a top user issue for the past 3 years.

Folders Enabled them to organize & quickly find their work

Bulk Delete Facilitated file management for multiple CUJs

Auto-populated Projects Facilitated organization & simplified project navigation

Users didn’t need:

Analytics Weren’t useful to most

Bulk Delete was a high user-value feature at risk of getting cut due to engineering complexity.
I met with the back-end engineer to pinpoint the issue.
I learned that the complexity arose from one step of the flow that was not crucial to the experience. To save the feature, I redesigned a simpler experience that was easier to implement.

Result: Users can now efficiently declutter their workspace with new multi-select tool + bulk delete.

My centralized search design required all of a user’s content (datasets, models, notebooks, and discussion posts) to be listed in one place. The tricky part: each content type had distinct formatting, metadata & interactions. I needed to standardize their formatting so that they looked uniform when listed together.

Users were excited at the prospect of selecting multiple files to delete at once. Our current design system, however, had no multi-select capabilities, and there were many deletion complexities to account for in my design.

Task: Enable users to quickly and safely delete items in bulk. The solution should:

Enable users to select multiple items at once
Warn of consequences of deleting items; give specialized warnings for each content type
Prompt users to review their selection before deleting

Task: Create one standard way to display 4 different types of content: datasets, notebooks, models, and discussion posts.

I started by auditing the metadata & format of each content type
I explored condensing content into 1, 2, or 3 lines
Ultimately, I chose the 3-line version in order to maintain consistency and include all necessary metadata

Result: With a standardized 3-line format, all content types can now be listed together, allowing users to efficiently view and search all of their work from one place.

Main page

Creating prototypes for testing

Concept Testing

The Goal

Enable content creators to efficiently organize, manage, and find their work from one centralized hub.

3) A Bulk Delete tool for efficient file management

User Feedback

2) Auto-generated project folders for assisted organization

Paper Wireframes

Explored structure & feature approaches

User Impact

With this new system, users can now:

Jump back into ongoing projects
Save time finding the right file
Organize their work into folders
Declutter their workspace

Design System Challenge: Standardization

Users loved:

Reprioritizing based on user feedback

After testing, I presented my user insights to the team and we aligned on the following prioritization of features:

P0: Manual folders for custom organization

P0: Bulk Delete for quick file management

P1: Automated Project Folders for assisted organization & project navigation

Removed from scope: Analytics

Interaction Challenge: Bulk Delete

🚩 Unforseen Constraints

As we approached handoff, we got a back-end engineer assigned to the project. He reviewed the design and informed us that the design required more back-end engineering resources than were available, and that we would likely need to cut some features to meet our launch timeline.

As the user advocate, I needed to determine which features we could cut and still give users a delightful and helpful experience. To make these decisions, I returned to my design goal statement, the PRD, and the strongest user feedback themes. Then, I worked closely with the engineers to understand implementation obstacles and find solutions that met user goals.

Adapting to New Technical Constraints

Reducing Engineering Complexity

Dark mode

Models tab

Projects tab

Next Steps

Metric Tracking: User Surveys - We expect to see a decrease in complaints about file management struggles in our yearly survey.
Metric Tracking: Content Creation Rates - One business goal of this project is to facilitate more content creation and sharing on Kaggle. We hope to see an increase in engagement and publishing numbers.
Future Opportunity: Analytics dashboard Although lightweight analytics weren’t useful to some users, others wanted in-depth performance analytics. Next up, I could partner with Research to explore user interest in an analytics dashboard.
Future Opportunity: Collaborative Project Folders During testing, users loved the idea of sharing collaborative project folders; however, that feature was cut from scope due to authorization complexities. Next up, I can partner with Engineering and Product to explore simpler ways to enable advanced project collaboration.

Final Designs & Next Steps

Lessons Learned

Include engineers throughout the design process to avoid unforeseen constraints and redesigns. Had I reviewed my designs with a back-end engineer earlier, I would have avoided redesigning an important CUJ at the last minute.

Test early whenever possible. Early user feedback helped me refocus the design on what users actually needed

Define the MVP early with cross-functional partners to avoid scope creep.

We did not have the resources to develop all of our proposed features. To determine what to prioritize in my designs, we needed to learn which proposed features would help users the most. Specifically, I needed feedback on the utility of four aspects of my design: metadata, folders, bulk actions, and performance analytics. To get answers, I traveled to an international Kaggle conference in Barcelona, where I independently conducted research with 15 Kaggle power users.

Defined research goals w/ PM • Prototyped 4 different designs • Developed research plan • Wrote question script •. Coordinated recruitment and logistics

How I Prepared

Design Phase 2:

Collaborating to save an MVP-Critical Feature