design & code, made by hand.

Built in SwiftUI with SwiftData. The first decision was where dictation lives. Most typing happens inside other apps, so Whisperly ships as a keyboard too: it types like a normal keyboard, and its mic key dictates into whatever app you are in — tap, speak, tap again.

That choice brings a hard moment with it. Dictating from a keyboard needs Full Access, and iOS shows a stern warning when you turn it on. Whisperly's answer is a plain explanation before you get there — what Full Access is used for, in the app's own words, without talking Apple's warning down. The same rule shaped every permission screen: say what it is for first, then one button that moves you on.

Writing for a reader like me changed the words everywhere. Short sentences, one idea each, no jargon, and a button that says what happens when you press it. Even the clean-up rules are written as sentences you can read and switch off one by one, instead of settings you have to decode.

The other rule was about trust. Whisperly is built not to swap or block a word you meant — anything a person could really say stays sayable. Its dictionary and learned corrections are there to learn your names and your spellings, not to overrule them.

Built in SwiftUI with SwiftData and zero third-party packages — no account, no developer server, no tracking. There's no public SDK for smart bottles like HidrateSpark, so instead of chasing hardware pairing the app reads what the bottle's own companion already writes to Apple Health. Zero setup, and manual logging works exactly the same without one.

The AI water buddy runs through Apple Intelligence, on-device by default; bring your own key and it only ever lives in the Keychain. Anything shared with an external provider is shown first in a byte-exact preview, and the buddy is written to never push you to drink more — goal changes always ask first, and can always be undone.

Apple rejected v1.0 three times — subscription fine print, completeness checks, metadata accuracy — and every round forced the app to be more honest about what it does and what it costs. Getting through App Review turned out to be a design discipline of its own, and the app shipped better for it.

The hardest design problem was building a game that knows when to stop cheering. Points cap at 150% of your goal, oversized entries ask for confirmation, and the buddy never says "drink more" — because a health app that only optimizes for engagement stops being healthy.

Built in SwiftUI with SwiftData for local-first storage — no account, no server, nothing leaves the phone unless you opt into your own AI key. The companion runs on Apple's on-device FoundationModels by default, with a scripted fallback where that's not available, and it's always disclosed as AI, never dressed up as a person.

The focus timer integrates with Live Activities, so a five-minute sprint keeps counting down in the Dynamic Island even after you leave the app. Between quests, a short transition ritual — breathing, a body scan, a glass of water — replaces the jarring jump from one task straight into the next.

Renaming a shipping app twice — Questling, then Focus Buddy, chosen only after both obvious names were already taken in Apple's registry — was a reminder that identity is downstream of constraints as often as it's downstream of intention.

Making the companion on-device by default turned out to be a design decision, not just a technical one. Once nothing has to leave the phone, every interaction can afford to be a little more honest, because there's no server-side incentive pulling it toward engagement over actually helping.

We went to a high school. Roskilde Highschool, Friday March 6th. Two team members, 9 groups of teens, face-to-face interviews. Phones out. We asked about climate change, campaigns, social media habits, and games.

Then we did something the brief didn't ask for: we borrowed VR headsets from KEA's tech lounge and ran usability testing with Greenpeace's 360° campaign videos. The teens loved it. The novelty pulled them in; the content kept them.

Real client briefs change everything. Research stops being theoretical; the people you're designing for are 13 and tell you exactly what they think. The target audience was younger than Greenpeace usually targets, which meant their existing campaign playbook didn't apply — we had to build something from scratch in two weeks.

Testing with VR was a gamble that paid off. It wouldn't have shown up in desk research as the "right" channel, but watching teens fight for the headset told us more than any survey could.

Full UCD process, by the book and then some. Business Model Canvas → personas → flow charts → wireframes → three rounds of style tiles → BERT (bipolar emotional response testing) on the style tiles → first prototype → expert review with the teachers → second iteration → think-aloud usability test → final.

The brand was developed from scratch — handwritten Wanderlust wordmark, backpacker silhouette mark, deep teal-to-coral sunset palette. Style tiles were tested with the target audience using BERT bipolar scales (Modern–Old Fashioned, Friendly–Intimidating, Adventurous–Safe). The winning direction informed everything downstream.

Group projects are 30% craft and 70% process. The Trello board, the shared Figma, the daily standups — none of that is glamorous, but it's what kept four people aligned across two weeks of iteration.

BERT testing changed how I think about visual design. We had three strong style tiles; without testing we'd have picked the one we personally liked. The audience picked a different one — and they were right for this audience.