BTIS • 2026 • Sole Product Designer • Shipped
Norbielink Design System
The design system behind every NorbieLink experience.
Built from scratch to help a team of one design, prototype, and ship enterprise software faster using AI-assisted workflows.
Overview
Designing the System Behind the Product
As NorbieLink expanded, consistency became increasingly difficult to maintain.
Without a shared design language, product decisions became harder to scale, interfaces became inconsistent, and AI-generated output varied across experiences.
I created a design system that serves as the foundation for both product design and AI-assisted development.
https://norbielink-design-system
https://norbielink-design-system
Impact
Scaling Product Development as a Team of One
To evaluate the success of the system, I focused on three outcomes:
• Increasing consistency across the platform
• Accelerating design and development workflows
• Creating reusable foundations for both humans and AI
47+
Reusable Components
2
Theme Modes
10X
Faster Iteration
Challenge
No System.
No Engineering Support.
When I joined the project, the product was evolving quickly but lacked a scalable design foundation.
There was no component library, no token architecture, and no shared rules for future development.
As the sole designer, I needed a way to scale product development without increasing complexity.
On Day One, the UI existed — but the system foundation didn’t.
Foundation
Building Rules Before Components
Before creating components, I focused on defining the system behind them.
Semantic tokens, accessibility requirements, spacing rules, naming conventions, and design principles became the foundation for every future decision.
Rather than designing screens first, I designed the rules that would generate those screens.
Before AI could generate interfaces reliably, it needed a system to follow.
Early AI-generated interfaces were fast, but inconsistent.
Components were used incorrectly, spacing varied between screens, and accessibility standards were often ignored.
The problem wasn't the model.
The problem was the lack of structure.
early Output by Claude - AI only gave me 70% right.
How I closed the 30% gap
I Don't Prompt The AI. I Engineer Its Context.
To close this gap, I stopped optimizing prompts and started engineering the system behind them.
Instead of telling AI what to generate, I defined the rules, constraints, and references that guided every generation.
Reliable generation came from structured system logic—not prompt length.
Final Output by Claude
Application
Applying the System in NorbieLink
The system became part of my daily design workflow.
Instead of creating every component manually, I used Claude to generate interfaces using the same tokens, rules, and constraints defined in the design system.
Each request followed a repeatable process:
Load Rules → Generate → Audit → Ship
three custom Claude Skills that act as the operating layer between AI and the design system.
CA0096497
Bayview Roofing LLC
Approved
Fees
BTIS Service Fee
$75
Broker Fee
$0
$0
Total Fees
$75
Payment Summary
Two Separate Charges
Both will be made to your card
Charge 1
Policy Fees
BTIS Service Fee
$75
Charge 2
Policy Premium
First Installment (6 mo)
$774
Total Premium
$774
Due Today (Your Card)
$849
Get Quote Now →
AI checks AI. I’m the editor, not the typist. Every NorbieLink component runs through the same loop — the system polices itself.
Production-Ready Output Generated Through The Design System Workflow.
What‘s next？
This project started as a design system. It evolved into an experiment in teaching AI how products are built.
The next step is expanding beyond visual systems and exploring how product knowledge, workflows, and decision-making can become part of the same shared context.
Up Next
Interested in how I navigated the challenge?
Hit me up for a deep dive into the project.