From a twin of myself to a product other designers can set up on their own. Designed, built and launched solo.
Role
Product, UX, UI, brand, code, copy.
Team
Just me, with AI
What this case study covers
Telltwin is a chatbot for designers’ portfolios that answers visitors’ questions in the designer’s own voice. I took it from a twin of myself to a public product with sign-up, onboarding, a dashboard and billing. It’s been live at telltwin.com since September 21. This case study covers the whole product and the reasoning behind it.
The problem
Hiring managers, recruiters and clients skim a portfolio. They have questions it doesn’t answer, and most of them won’t write to ask. They just leave, and you never learn who left or why.
Set it up once, and it keeps answering for you
Onboarding
Setup feels like a walkthrough by your own twin: you answer its questions to help it build itself, and everything you give it shows up on a card on the right.
No chat interface
I believe the onboarding is the crux of the whole project. I didn’t want a chat interface here. The experience had to stay controlled to get good results, close to the material you give it, and it still had to be engaging. I couldn’t let chat do all the work.
Answers that start themselves
Asking a bunch of questions, instead of keeping it really lean, raises the chance people drop off. So every question got small details that make it quicker to answer.
Under each question sit a few chips, drafted from what you’ve already given your twin, like your resume. Click one and it writes a follow-up for you, so you can answer by clicking instead of typing.
Pick a voice from a live sample
“How should your twin sound?” is two 3×3 grids: how you sound (formal to casual, bold to reserved) and how you explain (serious to playful, simple to expert). Every move rewrites the same sample line, so you compare tone and nothing else.
The dashboard
Never more than a message away
The dashboard is where you edit everything around your twin: your story, its voice, your links, what’s off-limits. The twin itself is the centerpiece, so its chat stays open whichever setting you’re in. Change its voice, ask it something, and see the difference straight away.
In the wild
The twin in the corner of this page is mine
Say a hiring manager has a question my case studies don’t answer. They can just ask, and it answers in my voice, from the material I gave it. Set it up once, and it keeps answering while you’re not around.
Go on, ask it something.
How I built it
One designer, up to 15 agents at once.
I designed and built Telltwin with Claude Code, with Codex reviewing the code. In Conductor, every change got its own workspace, agent and pull request, with up to 15 agents running at once.
6 weeks from the first line of the SaaS to production
950+ automated tests, run on every change
20 accounts since the September launch, most of them with a published twin
Go-to-market was part of the work too. I made the landing page, the launch films and a few ads.
Reflection
Today it feels that design decisions have nothing to do with pixels anymore.
Which model answers as your twin. Which faster one runs in the background, so the chips arrive while you’re still answering. A year ago I wouldn’t have called those design decisions. Now they very much are. And I think we’ll see less and less UI, as agents learn to talk to a product as easily as people do.
