Recent Work / Embr

"I don't always want to feel that wrecked"

That sentence came out of interview participant 3, halfway through a conversation about climate news. The full quote, in German, was blunter than anything we had written on a wall until then:

“Ich habe immer wieder Lust aufzuhören, irgendwelche Nachrichten mir anzuhören, weil ich mich nicht immer so abgefuckt fühlen will.”

I keep wanting to stop listening to the news, because I don't always want to feel that wrecked.

She was not uninformed. That was the thing. Every person we interviewed could explain the problem to us. Several could explain it better than we could. And almost none of them were doing anything, not because they doubted the science, but because the distance between what they knew and what they could personally do felt uncrossable.

Everything we built after that interview was aimed at that distance, and not at the knowledge.

My role. Three-person team. Mine: two of the seven interviews and their analysis, the wireframes, the coded prototype, the design system, and the edit and AI post-production of the product video. The concept decisions were made together.

A brief that started with a fire

The semester ran under the theme "Wehrhafte Natur", defensive nature, tied to Schloss Hardenberg in Velbert, a castle being converted into a museum. The brief had a real origin: during the exhibition planning, the roof burned down. Years of work and more than a million euros were suddenly unusable, and the client asked the agency the only question that mattered. What now?

That question became our assignment. Design a service around defence and adaptation.

We split the research three ways and each took one subtopic. Mine was the boundary between inside and outside, which sounds abstract until you compare a castle wall, tree bark and an exoskeleton and find the same structure in all three.

Protection does not come from a simple barrier. It comes from layered systems with controlled crossings, and those crossings are both the function and the weak point.

When we put the three subtopics next to each other, they met at one word: survival. That is where the project turned. The Neanderthal, a species that adapted to extreme environmental change and still, in the end, hit the limits of that adaptation, became our reference frame for how people handle climate change today.

The research phase ended on a question we could actually design against: how do you make someone's own capacity to adapt feel real to them?

Three archetypes, and the one we chose on purpose

Seven qualitative, guided interviews. Every participant got a polarity profile across ten opposing pairs, from anxiety-driven / capable of action to passive / activating. We laid the seven profiles on top of one another and read the overlaps.

Three behavioural patterns came out. Not demographics. Patterns.

We chose the Quiet Worriers, and that was a deliberate decision against the easier option. The Conscious Contributors would have loved anything we built and told us so. The Quiet Worriers are the group most likely to stay inactive, least likely to open our app twice, and hardest to move. They were also the only group where moving them would mean anything.

Everything downstream follows from that choice: the quiet tone, the small first steps, the refusal to alarm.

The thesis: a difference analysis, not a metaphor

The riskiest thing in this project is also its core. Comparing a person's climate anxiety to the Neanderthal sounds, said out loud, like a gimmick.

It is only a gimmick if it stays a metaphor. So we made it a measurement.

Embr does not say "you are like the Neanderthal". It takes one concrete trigger the user types in and places it structurally against the Neanderthal's situation, then reads the differences as leverage. A heat wave: early warning systems back then, zero. Early warning systems available to you in Frankfurt today, four. Currently used by you, zero. The gap between those last two numbers is not a comparison. It is a to-do list.

The point was never that we resemble them. The point is precisely everything we have that they did not, and how little of it we use.

That reframe is what turned a cultural reference into a mechanism. It also set the tone: the app has to be specific, local and measurable, because a difference analysis that stays global has nothing to show.

Six findings, six product rules

The deeper research question was why the Neanderthals died out, and what that says about handling uncertainty now. We ran it in parallel across several AI models, consolidated the outputs into one document, then verified every scientific claim against the actual papers by DOI. AI output counted as a step, never as a source. Over twenty verified publications ended up in the bibliography.

Six findings survived that process, and each one became a rule the product had to follow.

Read that list backwards and you get the product: no facts, small local steps, staggered by time, community attached, calm language. Nothing in Embr was chosen because it looked good in a wireframe.

The test that broke our entry point

Our original entry into the app was a share function: you see a post on Instagram that frightens you, you share it into Embr, Embr analyses it. Elegant on paper.

We tested it on a paper prototype with four people. The task was one sentence: you are scrolling Instagram, you hit a post that scares you, now share it with our app.

Nobody got there cleanly. Two problems surfaced at once. Sharing into an external app was not a mental model our testers had ready. And nobody could tell us whether a real user would even have Embr installed at the moment the fear hits, which is the entire premise of the interaction.

That result cost us our most distinctive feature as a first-run experience, and we had already presented it.

What we did instead of defending it: we shifted the design target from repeat users to first-time users, and built an onboarding in which the Neanderthal walks a new user through check-in, comparison and actions step by step. The share function stayed in the concept. It stopped being the front door.

A test that confirms you is a nice afternoon. A test that removes your best idea is the one that changes the product.

The other correction that semester was blunter. Our first concept greenlighting did not pass. We rebuilt the framing in a week and passed. It remains the most useful thing that happened to me in the project.

Building it: prototype, system, and a mixed-media film

The prototype is a plain web application. HTML, CSS and JavaScript, no framework, one HTML page with each screen as a container and a small JavaScript function switching between them, plus keyboard shortcuts so it survives a live presentation. Around twelve screens: check-in, a simulated analysis, the emotional read-back, the three comparison cards, the staggered actions, local initiatives on a map, a profile.

Underneath it sits a design system I built and then audited: terracotta and olive on warm white, Fraunces for headings, Inter for body, an 8px spacing base, three corner radii, a mobile-first grid with bottom navigation on phones and a left rail on desktop. Fewer type sizes than the first draft, deliberately, so the ones that remain are easy to tell apart in an anxious state.

For the Werkschau we produced a mixed-media product film over thirteen scenes. Lena, the persona, is a real person, filmed live. The Neanderthal is a 2D character composited onto the same footage. I cut the film and did the AI post-production.

Every generative asset in this project is logged: purpose, tool, model, timestamp, our own input material, the post-processing and how the output was used. Prompts included.

Documenting how we used AI took longer than using it. That documentation is the part I would defend in an interview.

What holds, and what does not

The honest state of this project: the mechanism is unproven.

We never ran a full user test on the finished prototype. We tested the paper version of one interaction, and that test earned us the onboarding. But the central claim, that comparing yourself to the Neanderthal strengthens your trust in your own ability to act, is still an assumption. If I could add one week to the semester, I would spend all of it on that single test with the Quiet Worriers, and nothing else.

Technically the same gap exists. The free-text analysis is simulated. There is no backend, so a profile does not survive a session, and the local initiatives are example data rather than real ones. The prototype demonstrates the flow honestly and does not pretend to be the product.

What I take with me is narrower and more useful than the app. This was the first project where I could trace every screen back to a piece of evidence, and where a test result cost me something I liked. Those two habits are the entire difference between a designed service and a decorated one.

I stopped asking whether a screen looks right and started asking which finding it came from. If there is no answer, the screen is decoration.