Launched in App Store
AI
Case Study
Timeline
Mar 2024 - Oct 2024
(25 Weeks)
My Role
Team
Outcome
MVP Launched
2nd Place at Mash-Up Demo Day
Background
People save useful links every day, from articles and videos to tools and references.
But saving does not always lead to revisiting. Over time, saved links become difficult to understand, organize, and find again.
Research
Survey Findings
A 160-person survey revealed that users were saving links frequently, but struggling to find and understand them later.
64%
had trouble remembering
what saved links were about.
26%
specifically struggled with
search-based retrieval.
The issue wasn't saving itself.
Users needed better context, organization, and retrieval.
Interview Findings
To understand why saved links were difficult to retrieve,
I conducted 5 one-on-one interviews with users who saved links through browser bookmarks.
“It's hard to determine if the link I'm looking for matches the title, so I have to open each link individually to check.”
Office worker in their 20s
Problem 1
Titles and thumbnails were not enough.
“If a link is important, I categorize it by type,
but it's too cumbersome to organize every link I save.”
Office worker in their 30s
Problem 2
Manual organization felt tedious.
PM
Mission
Strategy
Analytics
No serach results found
Q. Search for a recently saved link.
Link the interviewee was looking for
Interviewee's search keywords
"I couldn’t remember the content title, so I searched with keywords I remembered, but often couldn’t find what I needed."
20s job seeker
Problem 3
Search required exact keywords.
From Insights to Design Direction
Each finding shaped one core product direction for Linkit.
AI summaries and keywords provide context without requiring users to reopen every link.
AI recommends folders and categories based on link content, reducing manual sorting.
Content-based search helps users find saved links using remembered topics, keywords, or intent.
Design Solution
Feature 01 · AI Summary
Problem
Users often saved links quickly, then returned later to titles that no longer felt clear.
Design Decision
When a user pastes a link, Linkit reads the linked post and generates a short summary and keywords.
Instead of making users open the original page to understand the content, we brought the key information into the save flow.
Result
Users could quickly understand what a link was about without reopening it, using the AI-generated summary and keywords.
Feature 02 · AI Categorization
Problem
Manual folders and notes required too much effort to maintain.
Design Decision
We designed AI-powered category recommendations for links that were uncategorized or poorly organized.
Instead of asking users to sort each link one by one, Linkit groups related links together and lets users review the suggested organization.
Result
This reduced manual sorting while keeping saved links easy to browse by category.
Feature 03 · Content-based Search
Problem
Users remembered topics or keywords, but search results often depended on exact titles.
Design Decision
We expanded the search beyond exact titles by using AI-generated summaries and keywords.
Instead of relying only on title matching, Linkit searches across the content signals users are more likely to remember, such as keywords and summarized topics.
Result
Users could find saved links based on what the content was about, not just the exact title they had saved.
Launch & Recognition
MVP Launch · 2nd Place at Demo Day
Linkit was built and launched as an MVP through Mash-Up, a product-building community of 100+ designers and developers.
Our team presented the product at Mash-Up Demo Day and won 2nd Place among participating teams.
As an MVP built within a time-limited program, we did not yet have long-term usage data. If we continued the project, I would track return visits, category acceptance rate, and how often users reopened saved links after using AI summaries.
Reflection
What I learned
Launch day feedback showed that AI recommendations are only trusted when users can see the reasoning behind them. A feature can work technically and still fail if users don't understand why the system made a given choice.
What I would improve
