Elisabeth SaehrProduct designer, Lausanne
Symio · own product · 2026

From a blank repo to the App Store in eight days.

Symio is a symptom journal for people with a chronic illness. Log a bad day in a few taps, see whether a treatment changes anything, and bring a one-page summary to the doctor. I researched, designed and built it alone, with AI as my coding partner.

Role
Solo: research, product, UX/UI, code, store release
Timeline
6 Oct 2026, first commit → 1.0 on the App Store on 14 Oct → 1.4 on 10 Nov
Platforms
App Store · Google Play
Research
11-person survey · usability round with 7 people · 11 findings

In 30 seconds

8 daysfrom the first commit to the first App Store build. Apple approved 1.0 two days later.
11people with a chronic illness surveyed before deciding what to fix first.
9 of 11usability findings fixed in the app within a week. One rejected on principle.
0accounts, servers or trackers. Health data never leaves the phone.
Problem

People with a chronic illness answer their doctor’s precise questions from memory. Most only write symptoms down during a flare, if at all.

What I did

Designed a journal built for bad days, built it with Claude Code and Cursor, shipped it, then tested it with real people and fixed what they hit.

Where it stands

Live on the App Store and Google Play. No usage figures yet: the success targets are set, not measured.

The problem

It started with a friend who couldn’t answer her doctor

A friend lives with a chronic illness. During a new treatment, her doctor kept asking precise questions:

How many times a day?Since when?Worse before or after the new dose?

She had felt every symptom, but by appointment day the details were gone. That isn’t a memory problem. Chronic illness is tiring, symptoms blur together, and during a flare, writing things down is the last thing anyone does.

The person who knows the illness best often gives the doctor the least reliable account of it.

Existing apps make you choose. Bearable and CareClinic are complete but heavy to set up. Daylio is pleasant but not medical. Medisafe is about medication reminders. I wanted tracking light enough to keep up during a flare, and useful enough to pay off in the consulting room.

Research · survey

11 people, and the barrier I didn’t expect

An online survey from 15 to 17 October 2026, with people aged 21 to 32 living with lupus, endometriosis, coeliac disease, Ehlers-Danlos syndrome, asthma, migraine, PCOS, exocrine pancreatic insufficiency or a psychiatric condition. All agreed to the use of their answers. It is a signal from my own network, not a market study.

What stops you from noting symptoms more often?

It demoralises me

5/11

I forget

4/11

It’s not useful

4/11

The tool isn’t practical

3/11

Fatigue or pain · lack of time

2/11 each
What it changed

I had assumed effort was the main barrier. The most cited one is emotional. So the tone became a requirement: no streak guilt, no scores, a missed day is just a day.

How they track today
8/11rely on memory
7/11only note symptoms during a flare
7/11have never used a health app
5/11were asked by a doctor to keep a symptom diary
7/11take a treatment several times a day
10/11would give it at least a minute a day
4health professionals per person (median)
6 · 5iPhone · Android: reason enough to ship both

“The app wasn’t practical and I couldn’t see how it could help me.”

On why they stopped using a tracking app

“Trends, a link with new medications or doses, and a nice interface.”

On what other apps lack. Close to Symio’s pitch, word for word.

“Find the flares quickly, without looking through every day where nothing changed.”

On the one feature that would change their life
Key decisions

Three choices that shaped the product

My first complete design tested the idea well. Building it for real meant checking every screen against four principles: fast over complete, kind never guilt-inducing, show don’t interpret, private by design. Each decision below kept something people asked for out of the app, or put something in that the design didn’t have.

Decision 1

No account. The journal stays on the phone.

Chose
A local encrypted database, Face ID lock, export to CSV or JSON. No sign-up, no server, no analytics.
Rejected
Accounts and cloud sync, which my first design had: six screens of sign-up, email check and password.
Because
Health data is sensitive, and a sign-up wall is the worst start on a bad day. Privacy became the onboarding’s main message instead of a legal footnote.

First design · six screens removed

Six account screens from the first design, all crossed out: sign up, sign up with email, confirm your email, log in, create a password and delete your account.

Shipped

Onboarding screen ‘Garde ton journal privé’: data stays on the phone, nobody can access it, it is never shared or sold; optional Face ID lock and daily reminder. Screen ‘Supprimer toutes mes données’: a list of everything that will be erased, from first name to settings.

Onboarding: one screen, “Garde ton journal privé”: your data stays on your phone, nobody else can see it, Face ID and reminders are optional.

Deleting: one clear list of what will be erased, an offer to export first, and a typed confirmation.

First design → shipped. Six account screens replaced by one privacy screen and one clear deletion screen.
Decision 2

Numbers, never a verdict.

Chose
Before and after averages for each symptom, a dotted line where the treatment changed, “not logged” instead of 0. Words like “improves” are banned.
Rejected
“Pattern noticed” cards, automatic correlations and written summaries of the charts, although survey respondents and testers asked for them.
Because
Symio is a journal, not a medical device. “Nausea is fading, 62% lower” sounds kind, but it tells a patient their treatment works. The patient and the doctor decide what a difference of 0.7 points means.

First design

Home screen in the first design. Two cards are flagged: ‘Nausea is fading on Vedolizumab, 62% lower than your first week’ (judges the treatment) and ‘Pain tends to rise about 3 days before your period, so go gently’ (predicts symptoms).

Shipped

Shipped home screen: ‘Check-in commencé, Humeur : Bien, 3 symptômes notés’, the next appointment with its preparation, a prompt to add treatments and a neutral link to the reports.

Factual status: “Check-in commencé · Humeur : Bien · 3 symptômes notés”, the next visit with its preparation (0 of 4), and a neutral link to the Reports.

“Nausea is fading on Vedolizumab, 62% lower” tells the patient the treatment works. “Pain tends to rise 3 days before your period… go gently” predicts and advises.

Both sound kind, but both are medical judgements. Encouragement stayed, and it is only about consistency: “Plus que 2 jours avant ton 7e jour.”

First design → shipped. Two cards that judged a treatment and predicted symptoms were removed.
Decision 3 · after testing

Start from what hurts, not from mood.

Chose
A “Log a symptom” button on the home screen, and notes at any time of day. A day counts for the average of its values; “highest” keeps the real peak and its time.
Rejected
Rebuilding the daily check-in around symptoms, and a separate crisis button, which split logging in two and added a choice on the worst days.
Because
In testing, “How are you feeling today?” and its faces read as mood. People hesitated to log a stomach ache there, since you can feel fine and still hurt (PB-10). 4 of 7 also wanted to log more than once a day (PB-04).
Home screen in version 1.0: one entry point, the daily check-in.
1.0, tested. One way in: the daily check-in, framed as how you feel.
Home screen today: the check-in card, then a large ‘Noter un symptôme’ button asking whether anything changed since the check-in.
Now. “Log a symptom”, right under the check-in: has anything changed since?
Usability round 1

Testing the shipped app, and fixing it within a week

Once 1.0 was live, I tested it with real people at the end of October: three recorded remote sessions with a task script and a final questionnaire, plus four informal sessions where I watched people use the app. Every finding got a severity, a priority and a decision.

Build
Symio 1.0.0
People
7: three recorded on Lookback, outside the target group; about four informal, two of them with a chronic illness
Tasks
Onboarding, full check-in, a past flare, a new treatment, an appointment with a question, the doctor’s report, a treatment’s before and after
Data
Fictional persona and treatment, so no real health data was collected
Check-in: “so cool”Before / after: “very telling”P1: 5 / 5 on ease, clarity and wanting to use it
A tester setting up Symio at homeReal session, no sound
11 findings kept9 shipped as changes1 rejected on principle1 still open3 withdrawn: they didn’t match what I observed
FindingWhat people hitSeen byWhat changedStatus
PB-10 P1The home screen read as mood. People hesitated to log a stomach ache there.several, of 3“Log a symptom” on the home screen, notes during the dayShipped
PB-01 P1On Android, the keyboard covered the field you were typing in, from the first onboarding step.my testForms move above the keyboard on all 10 screens with a text fieldShipped
PB-04One check-in a day can’t follow symptoms that change from morning to night.4 / 7Several notes a day; average for the curves, the real peak kept with its timeShipped
PB-05Dose was a single field, overwritten on every change. You couldn’t see when it changed.3 / 7Dose history, a marker on the charts, before / after around any change, one line per dose in the PDFShipped
PB-02The intro illustration looked like buttons. People tapped it and nothing happened.3 / 7The card now plays the gesture itself, so it no longer looks tappableShipped
PB-03“Everything stays on your phone” didn’t reassure. Some thought Symio was online.≥ 2 / 7A fourth intro screen: no account, never sold or shared, journal encryptedShipped
PB-11Asking for a condition made people without a diagnosis wonder if Symio was for them.several, of 3First kept as is, then reversed: the step can be skipped, for people still waiting for a diagnosisShipped
PB-07No way to note waking up at night.2 / 7Night waking counter with a reason; people choose at onboarding what they track beyond symptomsShipped
PB-08The body map’s middle zone was called “Pelvis”, a word nobody uses for a stomach ache.my testRenamed “Lower belly”: smallest fix, same zone, past entries keep their placeShipped
PB-12Charts were hard to read alone. Some asked for a short written summary.several, of 3No change. A summary would interpret health data. Chart reading gets rechecked in round 2.Rejected
PB-06Periods are logged as a symptom, but lived as a cycle that gives context to the rest.2 / 7Left as is for now; the shape needs its own design workOpen
What didn’t work

Reaching the people I was designing for

Recruiting people with a chronic illness was the hardest part of the project. I posted in multiple patient groups and asked around many times. In the end, 11 people in my target group answered the survey, and the three recorded test sessions were with people outside it. Only two of my testers lived with a chronic illness, both in informal sessions.

The remote test tool made it worse. Participants switched back and forth between Lookback and Symio and lost the thread, some marked a task as done without realising there was a next step, and several worried that the microphone would record a private conversation.

So round 1 mostly tells me what confuses anyone, not what a tired person in a flare goes through. I said so in the synthesis, kept those counts as signals, and wrote down which findings need checking with the right people.

What I’d do differently
  • Start recruiting through patient associations on day one, in parallel with the build, not after the release.
  • Run moderated video calls instead of an unmoderated tool: fewer tasks, one instruction per screen, and a clear word about recording at the start. I’ve built the session kit for round 2.
  • Do the demo data import before the session, not during it.
How I built it

AI wrote the code. I made the calls.

I guided the AI from start to finish: every visual choice, and then, feature by feature, whether to build it at all.

I built Symio with Claude Code and Cursor. The product rules came first, in writing, and every change was checked against them before it went in.

  1. First commitProduct spec, principles, what V1 won’t do
  2. First App Store buildEight days in
  3. 1.0 approvedAfter a request for more information from App Review
  4. Survey11 people with a chronic illness
  5. Usability round 17 people, 11 findings, synthesis
  6. Fixes9 findings, each mocked up and validated first
  7. 1.4 builtFive versions in five weeks

What I decided

  • A mock-up before every visible change, compared before and after, then validated by me before any code.
  • The features, and which requests to turn down: correlations, verdicts, crisis button.
  • The words. “Appointment planner”, never “booking”: Symio doesn’t contact anyone.
  • The working rules: no network calls (a lint rule blocks them), no hard-coded text, each decision logged.

How I checked the AI

  • 962 automated tests passing on 9 November.
  • An AI code review (Greptile) made 5 comments. All were valid, all fixed.
  • A security audit found 1 high, 3 medium, 7 low issues. The high one is fixed and tested on a phone.
  • Real devices, every time. The Android keyboard bug (PB-01) passed every test and only showed on a phone.
The app

Built for a bad day first

Real captures of the shipped app with demo data. Symio is in French and uses the informal “tu”, to sound like a friend rather than a clinic.

Check-in, step one: five mood faces.
Check-in. One tap is enough on a bad day.
Body map with the chest selected at 3 out of 10.
Body map. Every zone at least 44 pt, read by VoiceOver.
A treatment page with before and after averages for each symptom.
Treatment. Before and after, stated as differences.
Preview of the one-page PDF summary for the doctor.
For the doctor. A PDF made on the phone.
An appointment with a preparation checklist and questions to ask.
Appointments. Questions ready before the visit.
Results so far

Shipped and tested. Not measured yet.

Symio has no meaningful usage data yet, so I won’t claim any. These are the targets I set before release, to measure in the next round:

< 10 sfor a full check-in
≥ 25%of people still logging after 30 days
Round 2with people in the target group, on the fixed version
What’s next

Ranked by what people told me

  1. Medication reminders, including “as needed” doses7 of 11 take a treatment several times a day
  2. Find difficult days fast in the historyOne respondent’s life-changing feature
  3. A one-tap “Nothing new” for stable daysRespects people who track only in flares, and keeps gaps visible to the doctor
  4. A home screen widget to log without opening the appAsked for in the survey
  5. Periods as a cycle (PB-06)Open finding from round 1

Only 1 of 11 would pay, under €3 a month. Symio stays free, with no ads and no data sold. How to fund it is still an open question.

What I learned

Three lessons

Say no to a request, but not to the need behind it.

Testers asked for written summaries, and I said no. The need underneath, reading the charts, stays on the list for round 2.

Designing for bad days makes the app better on every day.

A check-in you can finish during a flare is also one you’ll finish on a busy Tuesday.

Building fast makes research more urgent, not less.

AI got me to the store in eight days. Finding the right people to test with took longer than the build.

Symio · 2026 · All screens are real captures with demo data. Survey quotes translated from French. Next: CopyDash →