Yilin (Gevin) Zhu UX designer
Microsoft (Xbox) and Tencent internships, 2026 MSc UX Design, Wilfrid Laurier
I write the brief and the rules, and Claude does the volume work inside them.
Ordinal, CardCompare and Senior Mode were made this way; CardCompare began as a Figma Make prototype before that. Then I check what came back. When it is wrong, I say what is wrong and it goes around again.
Product direction, design decisions, the design system and the data policy are mine. Research legwork, data capture and code were done with Claude (AI) under that direction.
| What stays mine | What Claude does |
|---|---|
| Product directionCardCompare values Canadian credit cards on what the person reading actually spends. | Research legworkFinding and reading sources, such as a review of charts the FT has published. |
| Design decisionsDirect labels on Ordinal's charts. Three densities in Senior Mode, with only the senior surface enlarged. | Writing codeOrdinal's code, and CardCompare's rebuild in Claude Code since 6 Sep 2026. |
| The design systemThe tokens and components every Senior Mode screen has to follow. | Generating screens to specSenior Mode's 39 screens (27 senior, 8 adult child, 4 clinician), generated through Figma MCP from those tokens. |
| Data policyGaps stay blank, and a band counts at its middle in the composite. Ordinal takes a school into its table only once at least 3 core rankings list it: 1,247 in the build of 18 Oct 2026. | Data capturePlacements for Ordinal's 3,281 universities, in the build of 18 Oct 2026. |
| Final acceptanceI accept or turn down what comes back. On CardCompare, I made the final call on 21 items. | Running audit scriptsOrdinal's scale-honesty check and the script audit of Senior Mode's screens. |
One round starts with my brief and ends with my verdict. A no sends it around again.
What the page is for, and what it must never do.
Code, data or screens, made against the brief.
Screenshots at 1440, 768 and 390 pixels wide.
I write down what is wrong, or turn the attempt down.
The next attempt starts from my notes.
A redo goes back to the browser check.
Publishers print many placements as bands, such as 101–150. In the build of 18 Oct 2026, 3,699 of Ordinal's 9,051 published placements are bands. Ordinal prints each band the way the publisher gave it, and counts it at its middle in the composite.
Counting bands at the middle was my call.
I objected to the legend on Ordinal's charts. Claude then reviewed charts the FT has published: a legend appeared 0 times, and the plot area took 70.3% of the canvas. The legend came off, and each publisher's name now sits beside its own mark.
CardCompare started as a 13-page Figma Make prototype in May 2026 and has been rebuilt in code with Claude Code since 6 Sep 2026. Three redesigns of the card list row came back to me, and I rejected all three. The row went back to the original.
Senior Mode is a redesign exercise I did while interning in 2026. My mentor's verdict on the first version was that it looked AI-made. I insisted on looking at the real app first, and then threw out the original direction. The question of whether all three surfaces needed the senior treatment settled the new one: three densities, with only the senior surface enlarged.
They apply to my work and to Claude's alike.
When a ranking has no figure for a school, Ordinal leaves the cell empty and never fills it with an estimate. The rule began as my own request.
CardCompare's value and source counts are dated 7 Oct 2026. Ordinal's corpus counts come from the build of 18 Oct 2026, and the scale-honesty run from the build of 3,296 universities.
CardCompare has had no reader testing, and no page here says it has. Senior Mode's usability tests had 27 participants and found 4 issues, and all 4 were fixed. The testers were not older adults.
Senior Mode studies JD Health's public app. JD Health did not commission it, and it has not shipped. On CardCompare, the issuers' cards are what the site compares.
Claude's audit workflow shoots each page at 1440, 768 and 390 pixels wide, and the findings come to me to decide. Ordinal's Audit 8 still has 36 items waiting for that decision.
Counts from the scripts, audits and approvals behind each project.
125,248
renders checked by Ordinal's scale-honesty script, with 0 failures. Run by Claude on the build of 3,296 universities.
133
findings in Audit 8, 17–18 Oct 2026, which led to about 120 changes. Shot at 1440, 768 and 390 pixels wide.
18
cross-country mismatches between universities with the same name.
6,185
values from 946 source URLs. As of 7 Oct 2026.
1,833
CardCompare data corrections, each written only after I approved it.
38
cards missing their shadow layer.
10
greens that had drifted.
10
different button heights, from 41 to 62.
An architect stamps their drawings, which means they are responsible for them, and you don't stamp what you can't read. That's the line I'm trying to hold with AI.
Ordinal and CardCompare are live. Ordinal and Senior Mode each have a case study: Ordinal, Senior Mode. I'm open to UX and product design roles in Canada, remote or in the Greater Toronto Area.