Rothr mobilizes senior product designers to generate high-quality preference data, UI evaluations, and design rationale that teaches frontier models what good actually looks like.
We're building the infrastructure for design-domain AI training — through expert human judgment, structured evaluation environments, and scalable preference data pipelines.
Senior designers rank and explain AI-generated UI outputs to generate RLHF training pairs grounded in real design reasoning.
Structured evaluation environments where expert designers assess model outputs against accessibility, hierarchy, interaction, and craft standards.
Designers surface failure modes in AI design tools and agents — with the vocabulary to document why the output fails and how to fix it.
We built Rothr because we believe design expertise is the most underrepresented signal in the AI training economy.
The labs building the next generation of creative AI are racing to capture human preference data. The bottleneck isn't compute, and it isn't intent — it's access to people who can tell the difference between a UI that works and one that merely renders.
That gap is where Rothr lives.
We organize design intelligence — the judgment that comes from years of shipping enterprise products, navigating complex user flows, and knowing when something feels wrong before you can explain why. We turn that judgment into structured, scalable training signal for the labs that need it most.
As AI gets more capable, design expertise gets more valuable. We're building the company that makes that true.
The gap between output and quality is where your models are losing ground — and it's a data problem, not a compute problem.
General annotators describe what they see. Senior designers explain why it fails. The difference is the training signal your model actually needs.
Current platforms can't distinguish a 6-year enterprise UX designer from someone who completed an online course last month. You're getting noise.
By the time you've sourced, vetted, and onboarded design talent through generalist platforms, your training window has closed.
RLHF platforms weren't built for designers. Your domain doesn't map to their categories. Your expertise isn't captured.
Platforms that do accept designers pay crowd wages. Six years of UX reasoning is worth more than the market currently offers.
There's no vetted path from senior designer to AI training contributor. The demand exists. The on-ramp doesn't — until now.
A structured, reliable protocol for both frontier labs and design contributors.
Share your task type, design domain, volume, and timeline. Self-service or through our team.
We match and surface verified design experts against your specific requirements. No sourcing on your end.
Experts work inside your annotation tools or Rothr's structured task workspace. No migrations, no lock-in.
Start with a 5-person pilot. Scale to production teams of 50+ with the same vetting quality and structured output format.
Tell us your domain, seniority, and what kinds of tasks you want. No portfolio required.
We evaluate design judgment, not aesthetics. One test. No tricks.
Rothr surfaces relevant work from labs that fit your background and schedule. You decide what to accept.
Get paid $40–120/hr for async sessions of 30–90 minutes. Milestone-based. Transparent before you start.
Every task type is built around a consistent framework — so labs get comparable output across experts, and designers get work that actually uses their expertise.
01 Compare two AI-generated UI variants. Choose the stronger one and explain your reasoning in design terms.
02 Mark specific failures in hierarchy, visual weight, accessibility, and interaction clarity on wireframes or mocks.
03 Write structured design rationale that helps models learn what good looks like — and why.
04 Push AI design tools to their failure modes. Document what breaks, how it breaks, and what the right output should be.
For labs Filter your expert pool by the design domain closest to your model's use case.
For designers Your specialty is the signal. We match it to the labs that need it most.
Fast, structured, and highly calibrated expert signals built for immediate deployment.
Our research agenda focuses on one question: what does design expertise actually look like as structured training data — and how do we make it legible to frontier models?
Exploring failure patterns in current UI generation tools and documenting the quantitative delta between general human feedback and design expert preference models.
A standardized, public dataset of 5,000+ expert-scored interfaces designed to challenge the visual reasoning capabilities of multi-modal models.
Tell us what you need. We'll have a vetted shortlist in 48 hours.
Join 500+ vetted experts contributing to the frontier of AI.
