Healthcare · Diagnostics · 2025

Designing AI assisted experiences for preventive healthcare

Preventive healthcare AI companion — calm diagnostic experience

Diagnostic apps met anxious people with a catalog of hundreds of tests and a report full of red numbers. I designed a calm, profile aware companion that guides patient understanding instead of automating medical decisions, so people leave a result informed rather than alarmed.

Role
Experience Designer
Timeline
6 months
Team
Product, Engineering, Business
Focus
AI UX, Consumer health, Diagnostics
Overview

What if a diagnostic app calmed people down instead of winding them up?

Across interviews with everyday patients, one pattern held: people did not need more data, they needed to know what applied to them and what it meant. The catalog was complete, the reports were accurate, and still nobody felt guided.

Guide the choice
Replace the endless test catalog with a profile aware layer that surfaces only what is relevant to a person's symptoms and history.
Translate the report
Lead every result with a calm, plain language status, then reveal the numbers and trends underneath for those who want them.
Reveal the trend
Stitch single results into a longitudinal view so a marker moving over quarters is visible, not buried in separate PDFs.
Problem

The app answered every question except the one people were asking.

A patient arrived with a symptom and a worry. What they met was a promotional catalog of hundreds of packages, no way to tell which one applied to them, and a report that read like a spreadsheet of alarms. The tools were built to sell tests and print values, not to guide a nervous human through a decision.

The anxiety spiral
Hundreds of test packages No relevance to my symptoms
Book blindly, wait for a dense PDF
Red values read as alarms, upsells everywhere
Carry the anxiety into the doctor visit
Opportunity

Position AI as a thoughtful guide, not automation.

If the assistant explained instead of decided, the experience could change entirely: from a person guessing and worrying to a person who understands and acts with confidence.

Relevant, not exhaustive
Show the few tests that fit this person, with the physiological reason they matter.
Calm, not clinical noise
A neutral sentence first, the numbers second, so a result is understood before it is feared.
Longitudinal, not one off
Surface the trend across quarters so people notice change early, while it can still be acted on.
Solution

A calm, profile aware companion for everyday health.

One assistant that guides patient understanding across the journey: which test to take, what a result means, and how a marker is trending, all grounded in the person's own profile and history.

The measure of success was not the model's accuracy. It was a patient closing the app a little less anxious than they opened it.

Experiences

Two touchpoints, each resolving one moment of doubt.

The AI assistance concentrates where anxiety peaks: choosing a test, and reading a result. Each touchpoint answers a single question a worried person actually asks.

Contextual test recommendation interface
Choosing the right tests
A symptom aware layer recommends contextually relevant packages, each with a clear physiological rationale. Answers: which test actually applies to me?
Redesigned diagnostics summary with sparklines
Understandable reports
A redesigned summary leads with a calm natural language status card, then multi quarter trends underneath. Answers: is this okay, and which way is it moving?
Research

We listened to people at their most anxious.

Before designing a screen, I mapped how people actually experience a diagnostic journey. The same three needs surfaced again and again.

Context & relevance

"Tell me why this test matters to me, not just a list."

Interpretation

"I want simple understanding before visiting a doctor."

Proactive trends

"Help me notice long term trends earlier, rather than single data points."

Who we designed for
The Symptom Checker
Anxious, self diagnosing online

Needs to know which single test fits their symptoms and history, without a medical degree.

Frustration A catalog of hundreds of packages and no signal about what applies to them.

The Report Reader
Results in hand, doctor visit days away

Needs to understand a result in plain language before the appointment, calmly.

Frustration Red flags and dense tables read as alarms, so a normal result still feels like bad news.

The Longitudinal Tracker
Managing a marker over time

Needs to see how a value moves across quarters, not just today's number.

Frustration Every report is a standalone PDF, so the pattern that matters stays invisible.

Jobs to be done
Discover
Find the test that fits my symptoms and history in minutes.
Understand
Read my result in plain language before I see a doctor.
Track
See how a marker moves across quarters, not one value at a time.
Decide
Know whether a result needs action or just monitoring.
Trust
Get clinical guidance, not a promotional upsell.
Return
Come back for the next check without friction or dread.
Design Decisions

In healthcare, tone is a design decision.

The old experience was not a journey, it was a gauntlet of catalog, jargon, and upsell. The design goal was blunt: remove noise, and lead with calm.

Before / After
Before
The gauntlet
  • Scroll a catalog of hundreds of packages
  • Guess which test applies to you
  • Book, wait, receive a dense PDF
  • Read red values as alarms
  • Dodge promotional upsells
  • Carry the anxiety into the doctor visit
After
The guided path
  • Answer a few profile questions
  • See tests that match your symptoms and history
  • Receive a calm natural language summary
  • Read the trend, not a single scary number
Principles that shaped the AI
Confidence, not compliance
Build patient understanding and agency rather than automated reliance.
Clarity over intelligence
A diagnostic insight has to be comprehensible at a single glance.
Emotional sensitivity
Frame health markers neutrally to reduce cognitive and emotional load.
Explain, do not persuade
Trace every recommendation back to a concrete clinical rule the user can see.
Profile aware
Tailor guidance to age, history, and symptoms, never a generic catalog.
User controls depth
Lead with the plain sentence; let people open the numbers only if they want them.
Design goals in one frame Design goals: Explain — make health data simple and contextual; Calm — neutral clinical framing; Personalize — tailored to age and history; Control — user owns their detail depth.

Testing the first prototype, people read a normal result as bad news the moment a value turned red. The fix was not more data, it was tone: lead with a plain sentence, show the number second.

Outcome

From anxious catalog to confident companion.

+38%
Repeat diagnostic usage among chronic cohorts who need longitudinal, ongoing care.
< 6 hrs
Turnaround for digital summaries, down from the legacy 24 hour cycle.
AAA
Accessibility rating, the first diagnostic client optimized across all demographics.
Reflection

What I learned.

Calm is a feature.
In health, tone is not decoration. A neutral sentence placed before a red value changed how people felt about the exact same result.
Guidance beats automation.
The assistant earned trust by explaining, not deciding. Every recommendation traced back to a clinical rule the user could see, so agency stayed with the person.
What we didn't ship
  • Diagnosis or medical advice. Scoped hard to understanding and guidance, never clinical decisions.
  • An open ended symptom checker. Kept recommendations grounded in profile and history, not free form self diagnosis.
  • Gamified health scores. Deferred to keep the experience calm and clinical, not competitive.