Calibrate

Onboarding that reads your taste in 60 seconds, no form, no tour. It watches what you choose and tunes the feed before you've finished arriving.

Product Design|Onboarding|Personalization|Cross-Platform

Concept, design + build, solo·2026

Try the concept ↓

The problem

The first minute of a product is the one minute you can't get back, and most products spend it teaching.

A tour, a checklist, a "tell us about yourself" form with twelve fields, the new user is asked to do work before the product has done anything for them. For a utility, that's survivable. For a product built on taste, where the entire promise is "we get you", it's a contradiction. You can't open a relationship about shared sensibility by handing someone a form. A form is the opposite of being understood; it's being processed.

And taste is exactly the thing a form can't capture. Ask someone "what aesthetics do you like" and you get either a blank stare or a list of the tasteful-sounding answers they think they should give. People can't reliably describe their eye. But they can show it instantly; point them at two images and they'll know which one is theirs before they could explain why. The signal is in the choosing, not the telling.

So the real problem isn't "make onboarding shorter." It's: how do you make the first minute recognize the person instead of interrogating them, and do it in a way that feels like the product, not like a gate in front of it.

The approach

Don't ask. Watch.

Calibrate is a 60-second flow where you pick the image you're more drawn to, again and again, and the system reads the pattern as you go, building an aesthetic profile from your choices and writing it back to you in real time. By the end, the feed is already tuned. No form was filled. No tour was given. The onboarding was the product.

Choosing over describing.

Each step is a binary: this one or that one. No scales, no tags to self-apply, no vocabulary to learn. The interface asks for the one thing humans are fast and honest at, a gut pick.

Reading you back to you.

As you choose, the system names the aesthetic you're already expressing, "your eye gravitates toward the quiet and the organic." It's not recommending to you, it's articulating you. An audience that recoils from "AI recommendations" leans into "you, articulated."

Onboarding as first impression, not tax.

The flow is built to feel like the product at its best, fast, considered, a little poetic, so the first 60 seconds set the emotional register for everything after.

One flow, two surfaces.

The same logic, components, and tokens run on desktop and mobile. Designed once, deployed across, same recognition moment, native on each screen.

~60 sec

to a tuned profile

With zero form fields, the onboarding is the product, not a gate in front of it.

form fields

Built from binary picks, the one input people give fast and honestly.

hidden taste axes

Inferred from your choices, never shown as questions.

surfaces, one flow

Desktop + mobile sharing one set of components and tokens.

Try it

Pick the one you're drawn to. Watch it read you back.

This is the actual working prototype, real HTML/CSS/JS, not a click-through, running on desktop and mobile from the same code. Make a few choices and watch it read your taste back to you.

Where this breaks

Reading taste from choices instead of forms is the right instinct. Here's where the instinct runs out.

A 60-second read is a first impression, not a fingerprint.

A dozen binary picks capture a mood, maybe a lean, not the full, contradictory range of a real person's taste. The flow's confidence ("your eye gravitates toward…") is deliberately more certain than the data warrants. That's good theater and slightly bad epistemics, and I haven't squared the two.

Naming someone's taste can box them in.

"You, articulated" feels great when it's right and quietly limiting when it's close-but-not-quite, because now the feed leans into a version of you the product decided in the first minute. First impressions calcify, and there's no obvious moment in the flow where the person gets to say "actually, not that."

Binary picks smuggle in the curator's bias.

The taste the system "reads" is bounded entirely by the images I chose to show. Pick a narrow, tasteful set and everyone comes out looking like they share my eye. The flow feels like it's reading the user; some of what it's reading is me.

The four taste axes are machinery I'm not sure should be visible.

The current concept surfaces the axes during the flow. My instinct is that's exposing too much of the engine; the dials belong in a settings panel later, not in the poetic first-run. It's an unresolved tension between "show your work" and "keep the magic."

Recognition has a cold-start floor.

The first one or two picks have almost nothing to reason from, so the earliest moments are the weakest, exactly when a skeptical new user decides whether to keep going. The flow is most fragile at the precise point it most needs to land.

What I learned

I started out optimizing for short, fewer steps, less friction, and realized partway in that the goal was never speed, it was recognition; a 60-second flow that makes you feel seen beats a 10-second one that doesn't.

The hardest line to hold was confidence versus honesty: the moment the product names your taste is the moment it delights you, and it's also the moment it's most likely to be wrong, and I'm still not sure how much certainty I'm allowed to fake to earn the feeling.

And building the same flow on two surfaces at once taught me that "designed once, deployed across" is mostly a lie I tell about tokens; the recognition moment had to be re-felt on each screen, because the same logic lands completely differently in your palm than on your desk.

Try the live concept→

Explore more case studies

Design System