Enter the password to continue.
A private, voice-first app for documenting a difficult situation as it happens — a fifteen-second spoken note today becomes a dated, credible record months later, when memory alone isn't enough.
Daybook has to work for two very different people using the same seventeen screens.
Below is the walkthrough behind the design decisions — screen by screen, why each one is built the way it is, for both of them.
Checks in to record something that went well. Low stakes, low frequency — the app has to stay worth opening through months of quiet.
Recording a difficult situation as it unfolds, often late at night. High stakes, high urgency — the app has to keep him talking, not fill out a form.
Every screen, grouped by the moment it belongs to.
Setup is three optional fields and a visible Skip. No account, no email, no password wall. Someone who downloaded this on a lunch break can be recording eleven seconds later.
WhyEvery field here is a chance to abandon. The only reason setup exists at all is that a name makes the exported document read properly — so it asks, and lets you refuse.
The empty state names a concrete first move — something that went well this week — instead of showing a generic "No items" and leaving you to invent a use for the app.
WhyMaya downloaded this eleven months before she needs it. If day one feels pointless, there is no day two. Pointing at a win rather than an incident also sets the tone: this is not only for when things go wrong.
She taps record and talks: "Shipped the checkout redesign, conversion is up 12%, Priya said it was the cleanest handoff she had seen." No category picker, no date field, no title to invent.
WhyCapture first, ask questions after. The moment you put a form between the thought and the save, you lose the entries people make while walking to the car.
The recording is already saved before this screen appears. The AI work — date, people, entry type — happens after, and the line "Already saved, you can leave" says so plainly.
WhyIf someone force-quits during processing, they must not lose the entry. Saying so on screen is what makes the app feel trustworthy the first time it is slow.
Date, person and type are filled in and editable, and the entry closes itself out with a plain confirmation. Nothing is asked, because nothing more is needed.
WhyHow much the app asks scales with what is at stake. Interrogating someone about a good day is the fastest way to make logging good days feel like work.
Only now — with the entry already saved and visible above the ask — does the app raise signing in. Apple or Google, one tap, no password to invent or leak. "Not now" genuinely works and keeps everything local.
WhyYou asked for an account gate; the brief says capture must never sit behind a login wall. Both hold if the wall moves thirty seconds later. Daniel at 10pm meets a record button, not a sign-up form, and the data still ends up protected.
The entry takes its place on the timeline, tagged with the thread it belongs to. A single quiet line sets expectations for the pace — two or three a month, not a daily journal.
WhyNaming the realistic cadence protects against the other failure mode: someone deciding they are "behind" after a slow month and abandoning the app out of guilt.
Five weeks of silence. The nudge is soft, dismissible, and explicitly removes the pressure — "nothing has to be wrong for it to be worth writing down." Below it, the gap in the thread is drawn rather than hidden.
WhyThis is Maya's real risk: not friction, but silence. A nudge that implies she has failed at something gets the app deleted. A nudge that says the quiet stretch is normal gets an entry.
Every three months the app shows what it has, rather than asking for more. Four entries, three wins, two people — and a chance to add anything missed while it is still recent.
WhyThis is the answer to "what brings her back in month four." A nudge asks for effort; a recap gives something back first. It also surfaces the app's value at the exact moment she is most likely to doubt it.
Same button, same screen — but the copy adapts to the length: "Take your time. There is no length limit, and you do not have to make it make sense yet."
WhyDaniel abandons mid-entry if anything feels like a form. The only job of this screen is to keep him talking. Structure is the app's problem, not his, and telling him so out loud is the whole design.
A serious entry gets one or two follow-ups, phrased as offers rather than fields — "Was anyone else there who could speak to this?" — with Skip weighted equally and a promise that it ends soon.
WhyWitnesses and dates are exactly what a record needs and exactly what people forget. Asking at capture time gets a real answer; asking a year later gets a guess. Telling him how many remain is what stops it feeling like an interrogation.
Days later he adds a screenshot to the same entry. The privacy line under attachments states the boundary: files stay on the device unless exported, and are not scanned.
WhyEvidence arrives after the fact. If adding it later means creating a second, disconnected entry, the record fragments — and a fragmented record is worth much less in the room where it matters.
Six months of entries. The sequence — reported to HR, left off a kickoff nine days later, moved to weekend shifts five days after that — is legible because the dates are, and because the thread makes the order the primary reading.
WhyIn this context the sequence of dates is the evidence. Nobody remembers that the shift change came fourteen days after the HR meeting; the timeline does, because each entry was logged the day it happened.
Related entries group into a thread, and the thread writes itself a summary: how many times, over what span, and — the part nobody can hold in their head — the intervals. Nine days after the report. Five days after that.
WhyThis is the summary feature at the level where it earns its keep. Note what it never does: it counts and it dates, it never concludes. "Six occasions, two of them days after you raised something" is description. "This is retaliation" would be a legal opinion, and the app does not give those.
No trial, no countdown. Five entries are free with the full AI treatment, and the counter is visible from entry one so nobody is ambushed. Payment is asked for here — at Export, with eleven entries already built and the thing they came for one tap away.
WhyA three-day clock would expire before Maya had logged anything, since she records two or three times a month. Gating on usage instead means the ask lands at the highest-intent moment the app will ever have, and no free exports are given away in the meantime.
Six screens are shared by both people. Here the app asks the only question that genuinely differs: what is this document for.
WhySame data, two documents that are not variations of each other. Making the user choose the purpose rather than the format is what lets the app write each one properly.
Strictly dated, no grouping, no interpretation. Each entry notes when it was recorded relative to the event, who was present, and what is attached. It closes with a line stating plainly that it is a personal record and not legal advice.
Why"Recorded same day" is doing quiet, serious work: a contemporaneous record carries weight that a reconstruction does not. The disclaimer is not a legal formality, it is the product staying on the right side of a line it must not cross.
The same underlying entries, grouped into themes with impact stated in professional language. Twenty minutes of editing instead of a lost Saturday reconstructing the year from Slack.
WhyThis is why the app is worth keeping through the quiet months. The footer note — every claim traces back to something logged at the time — is what makes someone comfortable putting their name on it.
Rules that run through every screen above, not just the ones they're shown on.
No login, no form, no paywall between opening the app and recording. The account prompt only appears after the first entry is already saved.
Not optional, not nullable — absent from the schema entirely, so the data could never be searched by company name.
It can note the interval between two events. It can't call it retaliation. That line is enforced in the system prompt itself, not left to review.
The final mark took four full rounds: abstract "thread" motifs, a version that read as unintentionally anatomical, a too-subtle typographic exploration, and finally a literal open book — rebuilt twice more until a simplified two-shape construction held up legibly from a 16px favicon to full size. Every candidate was checked at four sizes side by side before being presented.
Thirteen years at HP, Nabo, and what's next — back on the main site.
← Back to the portfolio