WorkUp
UX Research & Product Design
Turning AI interview prep into practice students could trust.
I led UX research and the AI Interview redesign for WorkUp, a career platform with four AI-assisted job-search tools. Over 10 weeks, I guided a team of 15 designers through research, testing, and iteration to create a clearer path from practice to progress.
The product team adopted the redesign and presented it to investors, giving WorkUp a tested direction for its AI Interview experience.
Research participation
User testing
Error reduction
Task efficiency
23
13
~30%
1–2 min
students contributed to the research behind the AI Interview redesign.
participants validated the redesigned experience through usability testing.
fewer errors across the redesigned mobile and web journeys.
faster completion per task compared with the original flow.
Problem Space
More applications. Fewer offers.
The Class of 2024 submitted more applications than its recent predecessors while receiving fewer offers before graduation. For students searching for a first full-time role, every interview mattered.
27% fewer offers in one year.
WorkUp offered an AI Interview tool for that exact moment. But students had to navigate repeated setup, generic questions, and disconnected feedback before they could benefit from it.
Average job offers before graduation · NACE 2024 Student Survey
Discovery
Finding where confidence broke down.
We began with the existing product, then spoke with students to understand where WorkUp's experience diverged from what they needed.
The right idea, hidden behind the wrong experience.
A walkthrough of the existing AI Interview surfaced three problems before a single student was interviewed.
No Clear Way In
Overlapping paths and repeated setup screens made starting harder than it needed to be.
Practice From A Blank Page
Students had to create the questions they came to WorkUp for help with.
Feedback Without Direction
Scores, transcripts, and replay lived apart, with no clear next step after a session.
Understanding how students prepare.
We surveyed 48 students to learn how they practice, where they struggle, and how comfortable they are using AI for interview preparation.
89.6% Practice Alone
Most students rehearsed questions independently, without a consistent feedback loop.
68.1% New To AI Prep
WorkUp could not assume students understood or trusted the technology.
62.2% Struggle With Technical
The hardest questions extended beyond the behavioral prompts most tools emphasized.
Student survey
The feature students wanted to love.
Across 23 moderated interviews, students from different majors, year levels, and career goals moved through WorkUp while thinking aloud. The interview created useful pressure; everything around it made that value harder to reach.
Synthesis
Making sense of the mess.
Coded quotes from the interviews and survey clustered into two students who represented opposite starting points.
The interview worked. The experience around it did not.
Affinity map
Two students. Two starting points.
Erin needs enough guidance to begin something she has never done before. Maya needs practice specific enough to tell her whether she is ready.
User personas
The tool students wanted most was the one they trusted least.
WorkUp already had a realistic call format, skill ratings, and audio replay. Connecting those strengths through a clearer experience could turn a promising feature into practice students would return to.
How Might We
Help students begin with confidence, practice for the roles they want, and leave knowing what to improve next?
Ideation
Turning our findings into a direction.
Three design goals set the bar, then the flow, system, and sketches were built to meet it.
Three goals the redesign had to meet.
Card sorting set the structure.
We used card sorting to understand how students grouped WorkUp's interview content, running each session unsupervised to allow freedom of decision. Sequencing every participant's groupings gave us an ideal layout for the redesigned flow.
Card sorting
Sequenced results
One continuous practice loop.
Choose a set → set up the interview → preview the session → answer aloud → review each response → practice again.
User flow
Let the interface recede.
Clear hierarchy, restrained color, and focused components kept attention on the interview rather than the product around it.
Design system
Starting with structure before style.
Early sketches explored how students would enter the feature, move through setup, and leave a completed interview without reaching another dead end.
Low-fidelity sketches
Testing & Iteration
Refining the path from practice to progress.
We tested the first concepts with 13 students, then used repeated moments of hesitation to shape the next version. Drag a frame to move between versions.
One clear way into practice.
The first landing page still asked students to choose between practice sets and question sets, two names for nearly the same thing. We collapsed them into one library and let coaching point to what deserved attention next.
No more blank pages.
Creating a set felt suspiciously like doing WorkUp's job for it: every question had to be written by hand. Recommendations, filters, and search gave students a relevant starting point without taking away control.
An interview, not a meeting.
Invitations, scheduling, and session names made setup feel like arranging a video call. Committing to AI practice let us replace logistics with an interviewer choice and a preview of what was ahead.
Getting the interface out of the way.
The call screen made students think about recording when they needed to think about answering. We elevated the question, clarified the controls, and kept progress visible without competing for attention.
Results that go somewhere.
The first results screen had plenty to say and no idea what students should do with it. Bringing ratings, transcript notes, and replay together turned each weak area into the next practice session.
Final Delivery
A clearer way to practice under pressure.
The product team adopted the redesign and presented it to investors. More importantly, the final experience gave students a clearer way to begin, practice under pressure, understand their performance, and return with a purpose.
Final screens
Home & Coaching
A clearer place to begin.
Create direction
Organizing recommended, in-progress, and completed sets helps students understand where to begin.
Build momentum
Keeping unfinished sets within reach allows students to return without starting over.
Focus practice
Using past performance to recommend a skill helps students decide what to work on next.
Question Sets
Practice that matches the opportunity.
Reduce starting effort
Surfacing recommended questions helps students build a set without writing from scratch.
Increase relevance
Filtering by role, company, and difficulty helps students prepare for a real opportunity.
Support flexibility
Combining search with custom questions allows each set to reflect the student's needs.
AI Interview
Enough pressure to be useful.
Set expectations
Previewing the interviewer and session helps students understand what they are entering.
Create realism
A focused call environment recreates enough pressure to make practice useful.
Minimize distraction
Clear controls and visible progress let students concentrate on each answer.
Feedback & Progress
Results with somewhere to go.
Clarify performance
Pairing one overall score with scannable categories helps students understand how they did.
Connect evidence
Transcript notes and replay beside each rating show students what shaped the result.
Encourage return
Recommending practice from weaker areas gives every session a clear next step.
Next Steps
Opportunities for growth.
A few areas I would action if WorkUp were to continue its life cycle.
Reflection
Final WorkUp takeaways.
This project taught us that capable tools alone do not create trust. WorkUp already had four powerful AI features, but our research revealed how much students depended on guidance, personalization, and reliability to actually rely on them.
Leading with the problem.
It is easy to add features because they seem useful, but we learned to start from the problem and design only what solved a real, observed need. If an element did not map to something our research surfaced, it did not belong in the product.
Actions over answers.
Students would tell us one thing, then do something entirely different. Reading behavior over stated preference — watching where they hesitated, abandoned a tool, or worked around it — uncovered needs they could not articulate.
Collaboration is not one size fits all.
Each feature had its own team, pace, and problems, so I adapted how I worked with each one rather than forcing a single process. Meeting teams where they were kept everyone aligned.
The brief is not the user.
Early on we followed the direction we were given before testing it against real students. Taking an ambiguous brief too literally created flows that made sense to us but not to them. I learned to weigh the client's vision against the user's reality.
Thank You
It took a team.
A huge thank you to Tiffany and Edward, whose work as project managers made this opportunity possible for ACM and kept the whole team moving in the same direction. Thank you as well to Angelina and Jane, who I worked closest with, for making this project so much better.
Next case study