Zyltech, home
Email me
  • Not shipped
  • Mobile + clinician console
  • Health / accessibility

I kept one brand language at three densities and enlarged only the senior surface.

Senior Mode is a redesign exercise on JD Health's public app that I did during a UI design internship at Tencent in 2026. It is designed for older adults, their adult children and clinicians, in 39 screens on one token set. JD Health did not commission it, and it did not ship.

Senior Mode board, labelled in Chinese, setting the senior, adult child and clinician surfaces side by side. The senior phone screen has the largest type and controls, and the clinician view is a dense desktop dashboard.
Board 07, labelled in Chinese: the senior, adult child and clinician surfaces side by side, one brand at three densities. Open full size: board 07
Role
Product direction, design decisions, design system, review of the generated screens
Team
Me, with Claude (AI). My internship mentor reviewed the work and gave feedback.
Timeline
June to August 2026
Context
UI design internship, Tencent, 2026
Status
Not shipped. 39 static screens on one token set and drawn wireflows, with no clickable prototype.
Testing
Usability tests with 27 participants, as I report them. The testers were not older adults.
Scope
A redesign exercise on JD Health's public app. JD Health did not commission it, and nothing was delivered to them. The JD Health logo and the Joy mascot are JD brand assets, shown only as the subject of the exercise. The people, names and health readings in the screens are made up.
How it was made
Product direction, design decisions and the design system are mine. Research legwork, screen generation through Figma MCP and the audit scripts were done with Claude (AI) under that direction.

The problem

My internship brief was an AI health assistant for older adults. I took JD Health's public app as the subject because its home screen buries the way in for older users under feature grids and product shelves.

Bigger type suits only one of the three kinds of user the redesign had to serve.

Senior Mode board 02, in Chinese. On the left, a reconstructed home screen crowded with a sale banner, an icon grid and product listings. In the middle, the redesigned senior home with a greeting, four large entry points and the next medication. On the right, four before and after measurements.
The left phone is a reconstruction, not a screenshot: Claude rebuilt the current home screen at its real density after the new screens were finished. Board 02, labelled in Chinese. The middle phone is the new senior home at the same moment. Open full size: board 02

I kept the adult child and the clinician at their own density.

One enlarged layout for all three was on the table, and I turned it down. The adult child is an ordinary phone user and the clinician needs density, so senior rules would slow both of them down.

The brand stays fixed across the three surfaces and only the density changes. The older adult gets large type. The adult child's view keeps ordinary density, and the clinician works at clinical density, with abbreviations like qd and bid.

Clinician console in Chinese. Today's follow-up queue of 12 on the left, five of them in view; in the middle, one patient's readings, an AI case summary waiting for the doctor's confirmation and a 14-day blood pressure chart; on the right, long-term medication, safety checks and the doctor's orders.
The clinician console: today's follow-up queue, the patient record with an AI summary awaiting the doctor's confirmation, and long-term medication written as qd, bid and tid. Labels in Chinese. Open full size: clinician console

I derived the green from the parent brand and kept JD red to the logo and commercial slots.

I dropped an earlier dark-green direction after studying the real JD Health app. The primary is derived from the parent brand's green; white text on #0A8452 is about 4.7:1.

JD red appears only on the logo and in commercial slots, and purchase screens keep checkout in a bar pinned to the bottom. The page ground is a very pale green-grey instead of white.

Senior Mode board 03, in Chinese. Three panels move from JD Health's main app, through three conversion rules, to the senior version. Below them, swatches compare the main and senior greens, the page backgrounds, and the reds kept for the brand and for prices.
Board 03, labelled in Chinese: what carries over from JD Health's main app, the rules that change it, and the resulting greens, grounds and reds. Open full size: board 03

I gave anything the assistant writes its own colour and an AI label.

The competitive scan compared health apps on six moments, from finding the assistant to reading its answer. Each moment sets one app that handles it well beside one that does not.

In Senior Mode, everything the assistant writes sits in its own colour under a label that names it as AI and says what it drew on, so it cannot pass for a clinical reading. The chat header keeps a button that hands over to a human doctor.

Senior chat screen in Chinese. Two assistant messages sit in tinted bubbles, each under a purple AI label, with the Joy mascot as the assistant's avatar, a reminder offer below and a large hold-to-talk button at the bottom. A button at the top right switches to a human doctor.
Adult child's Today view in Chinese: the parent's blood pressure, blood sugar and heart rate, medication taken two of three times, and the assistant's daily report under an AI label with the mascot's avatar, with a button to call.
Left: the senior chat. Right: the adult child's Today view, with the assistant's daily report under its own AI label. Labels in Chinese.
Comparison board in Chinese for the assistant's entry screen. Outlined in green as the good example, iFlytek's health app gives its first screen to the AI assistant. Outlined in red, Ping An Good Doctor buries its AI entry under a doctor banner and an icon grid.
Research, moment 1, the way into the assistant. Green marks the app that does it well, red the one that does not. App Store screenshots, labelled in Chinese. Open full size: moment 1
Comparison board in Chinese for the AI answer page. Outlined in green, iFlytek's report reading splits results into flagged values and advice. Outlined in red, a general chat app answers in an unbroken block of text.
Research, moment 2, the AI answer: a report split into flagged values and advice, against an unbroken block of chat text. Open full size: moment 2

I kept the mascot out of serious moments.

Joy, JD's mascot, sits beside the assistant in chat. It does not appear on the emergency screen, which holds one press-and-hold control and the order in which the phone calls for help.

Emergency screen in Chinese with a large red press-and-hold button, a note that holding it for three seconds calls family in order, and a numbered call list: daughter, partner, then the 120 emergency line. No mascot appears.
The emergency screen. Holding the button for three seconds calls family in order, then the 120 emergency line. Labels in Chinese.

What changed along the way

  1. My internship mentor's verdict on the first version: it looked AI-made.

  2. I insisted on studying the real JD Health app first, and dropped the original dark-green direction.

  3. Whether all three surfaces needed the senior treatment became the question. They didn't, and the three densities came out of that.

  4. The second version went into a new Figma file.

  5. My mentor gave five points of feedback.

  6. Claude ran the consistency audit script across the screens.

Checks

Each number has a source or a date. The audit counts came from a script Claude ran.

Figma file, audit of 2 Aug 2026

  • 39

    screens, 27 senior, 8 adult child and 4 clinician, on one token set of 27 colour variables, 20 text styles and 3 elevations. The audit script Claude ran found the three rows below.

  • 38

    cards missing their shadow layer.

  • 10

    greens that had drifted apart.

  • 10

    button heights, from 41 to 62.

Primary green

  • 4.7:1

    white text on #0A8452. WCAG contrast formula, 4.73:1. Board 10 gives the same figure.

Reconstructed home screen

  • 13 to 19px

    body text.

  • 32 to 56px

    smallest tap target.

  • 3.1 to 7.4:1

    body text contrast. Measured on board 02. The before side is the reconstruction Claude built after the new screens were done, not a screenshot of the app.

Usability tests

27 people took part. They surfaced 4 issues, all fixed, and task completion rose by 21% from before the redesign to after it. All three figures are my own report of the round.

Gaps

  1. The testers were not older adults.

    The tests showed whether people could finish the tasks. Whether the type and contrast work for older eyes still rests on published guidance. The next round would be sessions with people in the target age range, on their own phones.

  2. The test write-up is thin.

    The counts under Checks are mine. The four issues, the tasks, the testers' ages, and whether +21% is percentage points or a relative change are not documented on this page.

  3. The before state is a reconstruction.

    Claude rebuilt the current home screen at its real density after the new screens were finished, and the before numbers were measured on it.

  4. There is no clickable prototype.

    The deliverable is static high-fidelity screens plus drawn wireflows.