Yeji Lee ✺ Product Designer

Compliance at the Speed of AI: One Retail Hub

Each retailers' compliance assessments were eating 100+ hours of brands' time. One Retail Hub redesigned that experience around AI, where trust, not automation, drives every interaction.

Contribution

Product strategy ✺ Research ✺ Design system ✺ Prototyping ✺ Frontend implementation

Team

1 Product designer (me) ✺ 1 Product Director ✺ 1 Project Manager ✺ 1 UX Researcher ✺ 4 full-stack engineers

Timeline

Nov 2025 - Ongoing

main image consisting of UIs from the casestudy

The Problem

Fashion brands spend 100+ hours responding to each retailers' compliance assessments, answering the same questions repeatedly, in different formats, for every retailer that asks. There was no shared system, no memory, no way to work smarter.

What I did

Sole designer, 0 to shipped in 3 months.

I created an AI-augmented design-to-development framework with engineering, eliminating handoff in favor of shared ownership: a design constitution, AI prototyping, and direct code commits. Design fidelity reached ~95%. Alongside that, I built the design system foundation and scaled it to cover AI behaviors and interactions: the human-in-the-loop patterns that define how the product feels and build trust.

An AI prototype built during early validation closed a major European RFP before production even started.

AI-Assisted Design-to-Dev Framework

The OneHub project became a proving ground for a new way of working. Traditional design handoffs used to lose 30-40% of design intent between Figma and production. I wanted to see if AI could close that gap without taking creative control away from the designer.

Traditional handoff 16+ weeks, 60-70% fidelity Designer Creates in Figma Handoff Intent gets lost Developer Interprets design Shipped 60-70% Weeks of iteration AI-assisted framework 4 weeks, 95% fidelity Constitution Shared rules Design Tokens + intent AI translate Figma to code Designer QC Edge cases Ship 95% Designer owns the full flow Engineer integrates AI translates, designer controls. No handoff erosion. 95%+ fidelity.

As a result, I developed a five-phase framework where AI acts as a translator, not a creator, which allows designers to stay in control from pixel to production code. A shared constitution constrains AI to execute decisions, not make them. Figma MCP translates design tokens directly to code. The designer QCs every component and designs edge cases directly in code, resulting in no handoff, no interpretation gap.

1. Constitution first

Designer + engineer + content designer define shared rules: design tokens, content standards, code quality. AI constrained by rules, not interpretation.

2. Design with intent

Designer sweats the details in Figma. Strict token usage, intentional naming, real content. Ideal state only, edge cases handled later in code.

3. AI as translator

Figma MCP reads design. AI checks against constitution. Asks 5-10 clarifying questions before writing code. "No interpretation" command prevents assumptions. Designer answers once, AI executes precisely.

4. Designer QC

Designer reviews token usage, component reuse, layout logic. Then adds what AI can't: loading states, error handling, empty states, micro-interactions, animation polish. All edge cases designed directly in code.

5. Engineer integration

Engineer reviews code quality, connects backend, adds auth and analytics. No design interpretation needed, all states already built.

The result: ~95% design fidelity in production (vs 60-70%), 4x faster delivery, and zero UX debt. After OneHub shipped, I documented the framework into a reusable guide and presented it org-wide, positioning design systems as critical infrastructure for AI-first product development at Trustrace.

Human-centered AI (HAI) Design Principle

AI in the passenger seat. One Retail Hub puts AI to work on the tedious parts, and keeps humans in charge of the parts that matter. Every interaction decision came back to the same question: when does surfacing AI actually help, and when should it step back?

That philosophy shaped not just the product interactions, but a new AI design system: intentionally understated, present when useful, invisible when not.

Outcome

Reduced brand compliance response time from 100+ hours to 36 minutes. Shipped in 3 months, presented at an OECD conference, and now live as TrusTrace's free-tier product.

View Product in Action

https://trustrace.com/solutions/one-retail-hub

Framework diagram connecting SRL cycle with AI roles

You may also like...

YouTube Shopping

Review Experience

AI + Learning

Fashion + Big Data

In-Car UX

One Retail Hub (You're here)