Soothe

Tools

• Swift
• SwiftUI
• SwiftData
• Core Haptics
• AVAudioEngine
• Claude Code

Soothe lets a commuter rest on the way home. A gentle pat, felt, heard and seen off one clock, says everything is fine until the stop is near. Then a clear, kind “Time to fly.” wakes them. It is a native iOS proof of concept: every screen is playable, and every ride, stop and signal is simulated. No location, no network, no real timetables. I made it as a product design case study in February 2020 and rebuilt it as an app in September 2026, with Claude Code doing the building one step per session.

Hear the pat

The ride

01/ 10

The next ride is the saved route leaving soonest.

A muted map. The underground stretch is drawn dotted.

The first minute is always calm.

The card shows the map or the timeline of stops.

Eyes closed keeps the screen on, since haptics only play in the foreground.

The copy says what the app is doing. Here it is simulated.

At the change the pat firms up and the orb turns warm.

Nearing, from the wake distance out: firmer and about 15% faster.

The wake runs until I'm awake, whatever the soft settings say.

Demo data, seeded in the app. No real rides.

Where it started

The 2020 brief asked: have you ever napped on the bus and woken up at the last stop? How do you wake a passenger just before their stop? I answered with saved routes, a tracking screen, statistics, reassuring notifications along the way and a vigorous alarm at the end. The persona was Sıla, an academic in İstanbul who worked overnight three times a week and slept on the second vehicle home, anxious every time.

In 2026 she became Noor, a night-shift nurse in Amsterdam-Noord who rides home exhausted and never lets herself sleep, afraid of the end of the line. A persona, not a research participant. Around her, the 2020 answer was taken apart:

• Routes, Track, Statistics and Settings became Today, Journey, Rest and You.
• Reassurance notifications became the Pat: haptic, sound and light on one clock.
• A pulsate rhythm and white noise became three rhythms, Heartbeat, Pat and Tide, and three ambiences, Rain on the canal, IJ wind and Brown noise.
• The vigorous wake alert became Nearing, firmer a few stops out, and then Arriving: “Time to fly.”, until “I'm awake”.
• Badges became Moments, gentle milestones that are never gamified.

1. Brainstorming, phase 1

   43 notes

   • Feelings13

     • Boring
     • Lethargic
     • Nausea
     • Tiresome
     • Depressive
     • Claustrophobic
     • Comfortable
     • Negative
     • Mechanical
     • Anxiety
     • Mundane
     • Familiar
     • Unfamiliar

   • About Solution12

     • Childhood
     • Multi-Platform Smartwatch
     • Desmond Morris
     • Process Progress
     • Maternal Patting Reassurance
     • Chirping
     • Haptic Feedback
     • Womb
     • Headphones
     • Cradle
     • Multi-vehicular Situations
     • Multi-sensory reassurance at every stoptyped

   • About Travel18

     • A Pack of People
     • Dark and Dim
     • Dusk
     • (Sleeping on bus) Effective
     • (Corporate Service) Grouping
     • Cars
     • Complicated
     • Range
     • (Corporate Service) Hit or Miss
     • Read the Plate
     • Alot Start
     • Crowded
     • Possible
     • System
     • Anxiety
     • Busy Mind
     • Concrete
     • Awareness

2. 

3. Main user flow

   User context

   User gets out at the usual hour, walks to the stop she has been gotten on the bus.

   Tech context

   User's device has adequate battery power. Cell coverage is abundant for the first phase.

   1. Phase 1, bus

      1. 01

         App starts to listen geolocation due to the time pattern of the user.

         • Clock data initiates
      2. 02

         App taps into the location sensors and confirms the location is start point of the commute.

         • Location data confirmed
      3. 03

         According to the automation rules defined by the user (user opted for sound, haptic & white noise) starts as soon as user connects headphones.

         • Listening location data
      4. 04

         Throughout the first phase of the commute, user hears slow, rhythmic sound and feels haptic feedback simultaneously & synchronized.

         • Listening location data
      5. 05

         User gets close to the stop to get off the bus and take a train. Now app grasp the situation through location services and changes auditory and haptic feedback to a vigorous rhythm to notify user: time to get off!

         • Location data confirmed
         • Clock data registers

   2. Phase 2, train

      6. 06

         User gets into the subway station. Cell coverage lost. App transitions to accelerometer mode.

         • Location data N/A
         • Cellular signal lost
      7. 07

         By listening accelerometer data, app grasps train's initiation of movement.

         • Listening accelerometer data
      8. 08

         App counts the stops of the subway utilizing accelerometer data. Throughout the commute, app reinitiates soothing feedback until it is time to get off.

         • Listening accelerometer data
      9. 09

         User gets close to the stop she needs to exit, this time using accelerometer data and confirm from clock; changes auditory and haptic feedback to a vigorous rhythm to notify user: time to get off again!

         • Accelerometer data initiates
         • Clock data confirmed
      10. 10

          As soon as there is location data to confirm the commute is over, app disengages and registers commute to be found in statistics.

          • Location data confirmed

4. 

5. 

6. 

What commuters told me

A short questionnaire in 2020 asked about the commute to work and back. For three words on it, most people wrote negative ones: boring, tiring, prolonged. Sleeping through it makes sense. Four findings shaped the app, each with its 2026 answer under it. The board printed no sample size, so I give none.

A pat, not an alarm

The idea came from Desmond Morris. In Intimate Behaviour he describes how a mother's repeated pat tells a baby there is no danger, the same kind of signal a bird gives with its head before it flies. “The repeated patting is therefore doubly soothing.” (Morris, D. (1997), Intimate Behaviour, New York: Kodansha America, p. 24.)

My 2020 sketch notes turned that into one line: the app pats you from time to time to say everything is OK, until it is time to fly. What changed in 2026 is what happens then. In 2020 it warned you vigorously. Now it says “Time to fly.” and waits for “I'm awake”.

The 2020 flow asked for sound and haptics “simultaneously and synchronized”. In the app one clock emits the beats, and haptics, sound and the orb follow the same emission, so they cannot drift. The sound is made in code, so there are no audio files. The phases change the feel, not only the words: resting is soft, nearing is firmer and about 15% faster, and arriving is a rising pulse with two knocks every two seconds until the rider taps “I'm awake”. The wake ignores the rider's soft settings on purpose. Hear it plays one bar here, made the way the app makes it.

Two sheets change the feel mid-ride. Haptics can only be felt on a real iPhone, and I have not verified them on one yet.

On the first run the pat is felt before the first ride, from the second card.

First run

01/ 04

The ride home

The demo ride is Home after the night shift: metro 52 from Zuid to Noord under the IJ, a four-minute change, then bus 34 to Grootzeil. Twenty-six minutes, one change, ten stops. Three rules place the phases. The first minute is always calm. Nearing starts when the stops to go reach the rider's wake distance, one to three stops. Arriving starts 45 seconds before the destination, and only “I'm awake” ends it. Underground, the copy says what the app is doing: “Underground. Counting stops by motion.” In this build that is simulated too.

Calm, on purpose

The design system is a set of refusals. Matte paper, one primary action per screen, one slow curve with no bounce. No gradients, no drop shadows. Liquid Glass only on the two controls that float over the map. New York for display, SF Pro for the interface, Dynamic Type text styles only. If a screen feels busy, something is removed.

Contrast is a test, not an intention: ink at least 7:1 on paper, secondary text and line chips at least 4.5:1. Two values failed when measured and were changed. Both are on the palette below.

Light and dark on every screen, both from the same tokens.

Rest counts the minutes a rider actually rested: settling, resting and underground, never the change or the wake. Moments are noticed, never scored: a first ride, a first rest underground, ten rides. Every ride shown here is seeded demo data, and each screen says so.

Accessibility was part of done: VoiceOver throughout, phase changes announced, Reduce Motion turning pulses into opacity, and Eyes closed keeping the screen on because haptics only play in the foreground.

What is simulated

• Every ride. No location, no sensors. A pure ride model lays the route out in ride time, so a ride lasts exactly the minutes on its card, and a test runs one at 6000× in under a second.
• Underground. The app counts stops by motion in the copy only.
• Amsterdam. Station names and line numbers only. Line order checked against GVB's 2026 network, stops from OpenStreetMap, durations computed. No operator is involved.
• Two switches. Let Soothe learn it and Start with headphones are mocks, and their captions say so.
• The history. Seeded into an empty store so the charts are never empty. It is not user data.
• Haptics are not verified on hardware. The Live Activity and the Siri intent have not been checked by eye.
• No one has used it. No users, no store listing, no research since 2020.

How it was built

Claude Code built it, one step per session, over two days in September 2026. AGENTS.md is the contract every session reads first: the non-negotiables, the workflow, and a router that names the one doc to read for the area being touched. The kickoff prompt is one line: read AGENTS.md and do the next step. A step is done only when it builds, its tests pass and its docs are current. Then the agent stops. It never commits its own step. Starting the next session is my approval of the last one.

• S0. Scaffold: the agent docs, the XcodeGen project, a Makefile, the icon and an empty three-tab shell.
• S1. Design system: tokens, components and a hidden gallery.
• S2. Amsterdam data and routes: the catalog, SwiftData, the seeder, Today, route detail and New route.
• S3. Feel engine: the rhythm clock, haptics, the procedural soundscape, the Sound and Touch sheets.
• S4. Journey: the simulator, map and timeline, the orb, the change, underground and the wake.
• S5. After the ride: the summary, Rest, Moments and seeded history.
• S6. Onboarding, You, the Live Activity with Dynamic Island, and the Start my ride home intent.
• S7. Polish: dark mode, Dynamic Type, VoiceOver, Reduce Motion, empty states, copy and screenshots.

Apple frameworks only, no third-party dependencies: 133 Swift files, about 7,050 lines of app code and 2,290 of tests, 205 tests passing, no warnings.

This page was made the same way, one step per session, and I read every word before it landed.