A design system, an AI workflow, and every core flow rebuilt on it

Design and engineering had no shared source of truth. Components drifted between what was designed and what shipped, patterns got re-argued on every feature, and discovery ran sequentially: about three weeks of research, then three of design.

On top of that the platform had to scale across web, chatbot, mobile web and native Android and iOS without every squad reinventing basic UI.

I went through 60+ existing Figma files and pulled the scattered patterns into one library of about 50 components covering the platform's interfaces. The foundations (spacing, typography, sizing, motion) went in as a variables and tokens architecture applied straight to the components, so the system holds up across every surface.

A full re-architecture of the front and back end meant home, PLP, PDP, cart, minicart, checkout, authentication and my-account all had to be realigned at once. I ran usability testing on the existing onboarding to find where and why people abandoned registration, not just that they did, then iterated in testable increments instead of one big redesign.

The bigger question was the workflow: how to get discovery from six weeks to two or three. I worked on the agent architecture and on finding the flows that actually saved time. Research agents handle market study, benchmarking and data analysis. AI tools cover rapid prototyping and validation. A Figma-to-specs agent cut manual hi-fi work from a week to about a day, and the handoff carries the full discovery context into the development pipeline.

I led a shifting team of mixed seniority, aligned expectations with product and the client, and gave the designers ownership. I coached them on design systems so they could maintain it too, instead of the whole thing depending on one author.