TLDR
In six months I designed an app for expecting mothers across three stages: planning, 40 weeks and the first months after birth. Inside: a week guide, a daily routine, a mood check-in, a meal scan and an AI chat. The early-stage startup found me through LinkedIn.
After three rebuilds I left, and the product never shipped. Three months later the founder compared my version with the next designer’s work and came back to mine as the base for the product.
What was on me
- Built the product model. Scenarios, priorities and feature logic: the vision was only a draft when I arrived.
- Laid out the information architecture. The week became the unit of content and navigation: 40 weeks and six phases.
- Designed the core of the product. Five sections and the “Am I ok?” triage with three calm outcomes.
- Split rules from AI. Rules assemble 276 day-plan combinations; AI stays where there is a dialogue.
- Set the visual language. “A calm companion” for two markets: Russia and the US.
One app for planning, pregnancy and postpartum
The audience is in an anxious period: the product’s main answer is “this is normal”, and I set the tone and density of the screens for it. On the market the cycle lives in one app and pregnancy in another; only three leaders out of 13 cover all three stages. So I put them into one product.
What 13 pregnancy apps already have
Decisions rest on a survey of five pregnant women and a teardown of 13 apps from the US and Russia. I asked each app the same questions and put them in one table: 141 captioned screens, 35 sources.
A “this is normal” list and a daily routine I found nowhere.
Today answers “what do I have today”
The header holds the whole context without a transition. Below it, the day itself: the routine, the mood check-in, the week’s articles, the affirmation.
The week — the unit of content and navigation
The week opens where the question about it comes up: the tile on Today leads to “Your journey”, and from there to the week guide. The guide answers with three lists, before any chat; I wrote the weekly texts in advance, and the personalisation lives in the logic.
What’s normal
What you may notice
What to see a doctor about
Routine — soft habits
Morning and evening, required and optional, progress x/5, without guilt for what’s left undone. “Adjust today’s routine” drops the optional items in one tap, and the counter recounts against the lighter day.
The focus and the habits
A task, opened
Mood is part of the core
A daily check-in with flower characters and a free-form note: five states, a scale with no wrong answer.
The meal scan answers in words, no scores
The scan shows lunch as a contribution to the week, with no scores or percentages: first whether the dish is safe, then the focus for this stage. At week 18 that is protein, iron and calcium; a rule picks the focus by week, AI takes no part.
The price of the decision. An instant verdict on a single plate, Cal AI’s main mechanic, was dropped: for an anxious audience a verdict on one plate is shame and false precision.
History shows what happened, with no count of what you missed
History opens from the calendar in the Today header and runs on weeks, like the guide. A day shows five rows the woman entered herself: symptoms, mood, routine, note, scan.
The cost of the decision. Of the five marks in the week strip I kept three: a day with a record, an empty day, the opened day. The fill doesn’t show how much is filled in: a ratio would turn the week into a report card and an empty day into a miss.
An affirmation at the end of the day
The affirmation sits at the very bottom of Today and opens on a tap: until then you see a blurred patch and “Tap to reveal”. The routine above counts “1 / 5 done”; nothing in the affirmation is counted.
What it stands on. Self-affirmation theory (Steele, 1988): affirming your own values lowers the defensive response to threatening information, and symptoms and “what to see a doctor about” sit right next door. So the affirmation is opened by her own tap, when she is ready.
Rules build the day plan, AI answers questions
- Rules: the daily plan without generation. (40 weeks + 6 phases) × 6 profiles = 276 combinations; the day’s card is Body, Food, Mind, Red flag.
- AI: the “Am I ok?” triage. The situation in her own words, 1–3 clarifying questions and one of three outcomes: “probably ok”, “worth checking”, “don’t wait”.
- Boundaries: the chat doesn’t replace a doctor. Safety rests on the architecture, with no reliance on the text: rules and three outcomes.
The assistant speaks first: the app already knows the day, the week and the closed tasks. The pause before the answer shows three sources (the week, the day, her own entries) and no chain of thought: reasoning out loud about a symptom reads as a diagnosis.
From a metrics tracker to a companion for her state
1.0 was made before me, then three rebuilds of mine, and each one shed a layer of counters. First the week became the unit of content, then the day stopped being a percentage, then the woman’s own state moved out of “metrics” into the core.
1.0, before me. Percent and counters
2.0. The day as Body / Food / Mind cards
Now. One counter, about the routine
The work they came back to
Three months after I left, the founder compared my version with the next designer’s work, came back to mine and asked me to return.
I did not. Shipping a product like this takes full-time work; evenings would not get it there. Meanwhile the founder kept adding features, and the MVP moved further away.
I built five metrics into the design. Weekly return, days with logged meals and with the affirmation opened, retention D7/D30, questions answered by “this is normal”.
What I took with me
I pin the product scope down at the very start and explain it to the founder through what matters to him: time and risk. Towards the end of my work he wrote a product deck himself, the vision the project had been missing.