Work in progress. This is a live project conducted as part of the Carnegie Mellon Master in Human Computer Interaction Program. Design is still moving, user testing is still being completed and things may change as results come in.
I lead design on the prototyping work for RoboticsCareer.org, the ARM Institute's platform connecting people to careers in robotics and advanced manufacturing. ARM brought our CMU team in with two goals: grow the platform's reach, around 100,000 annual visitors today with ambitions of a million, and answer the question it couldn't, what actually happens to the people who use it. Our answer is a home base users return to for their whole career, and every return is an outcome ARM can finally see.
Client
ARM Institute
MHCI Capstone Project
My role
Lead Product Designer
Timeline
Jan 2026 – present
Skills
Product design, User research & testing, Design systems, Frontend prototyping, AI-assisted development
Team
Stephen Chen · Kayla Windust · Rachael Shivam · Tian Zhou
Problem: The platform goes blind at the moment it succeeds
ARM wants to report real outcomes, who got trained and who got hired, to the stakeholders funding the platform and the users deciding whether to trust it. But the moment someone clicks apply, the site hands them to an external listing and never hears from them again. And when I dug into the account data, the story got starker: 87.3% of the 4,034 people who ever made an account never returned after their first session. The platform can't prove value it never sees, and it can't grow on users who never come back.
Research: What was keeping people from coming back
To get at what was driving this, we spent four months on a seven-method research program, from high school workshops to platform analytics. Three findings set our design direction
Young people don’t reject the work, they just can’t picture themselves in it.
When we asked high schoolers whether a robotics role could be for someone like them, the most common answer was "maybe," never for identity reasons, just that nothing connected the work to their own interests.
The site loses people before it shows its value.
Our usability work catalogued 82 issues, and testers asked out loud why they'd use RoboticsCareer.org at all if applying just lands them on LinkedIn.
The upside lives behind the login.
Logged-in users engage at around 95% on desktop and mobile alike. The people the platform wins, it wins hard. It just wasn’t winning many.
So we reframed with ARM: stop chasing the upstream problem of making people care about manufacturing, make the platform as good as possible for everyone who lands on it, and finally give ARM a way to see outcomes
Solution: A home base for a whole career
Our answer gives the platform a center of gravity it never had: a logged-in home base where someone can plan their entire career, from early exploration through every new job, and where every visit quietly feeds ARM the outcome data it's been missing. We designed a quiz that pulls young people in, a dashboard that brings them back, and, in parallel, a redesign of the live site's most trust-breaking surfaces
Customizable career dashboard
Self-reported outcome tracking
Narrative interest quiz
One design system across Figma and code
A dashboard that opens on your progress.
The center of the work is a new logged-in home I designed and built end to end. A returning user opens on their own momentum: applications, trainings, tasks, what to do next. It fills a gap other career trackers leave open, every part of a career in one place, the job hunt and the training that leads to it tracked side by side.
The worst moment becomes the reason to return.
When a user clicks out to apply, the dashboard logs it. On their next visit a small check-in asks how it went, and one tap updates the status. That gives users the tracker they'd otherwise keep in a spreadsheet, and it turns the platform's blind spot into its strongest return loop.
A quiz you play, with results you can trust.
The interest quiz is the front door for younger users, built for the gap our research found: a playable day in the life that connects your interests to one of ARM's three robotics roles. The craft lives in the results: every match shows its work, what you chose, what you passed on, and why it adds up, transparency that earns the trust a score alone never could.
Built at AI speed, with production discipline.
To move at the pace this project demanded, I built agentic design workflows around every prototype: a shared token package feeding Figma and three codebases, agent skills that carry a change from Figma into code and back on a single command, and decision logs and design rubrics that agents enforce, all of it steered by the user-centered judgment my training and experience afford.
Impact: Where things stand
After our spring research presentation, the director of the MHCI program called this "not only the most well researched ARM capstone project, but the most well researched project I have seen in MHCI."
The work is now in its closing stretch: final polish on every prototype, a last round of user testing that quantifies how the designs improve on the live site, and implementation documentation for ARM's dev team. This section, and the case study as a whole, will be updated when the project wraps.
Go deeper
Turning a blind spot into a retention loop
Designing the apply handoff: auto-logging, one-tap check-ins, and a status system with no red in it.
Two quizzes walked in, one walked out
A narrative quiz and an assessment went head-to-head. Users trusted one and enjoyed the other, so I built both into one.
How one designer ships three codebases
The agentic workflow behind the build: slash-command Figma-to-code sync, decision logs, rubric agents, and test gates.