← Case studies
×
Surge Intelligence
Fin · 2026
Situation
Design at Fin scaled through people: more designers, more reviews. With agents doing more of the build and R&D moving faster than any design team could review, the quality bar depended on who you happened to have access to ask.
Task
Make the quality bar part of the infrastructure instead of the headcount: encode Surge — its components, patterns, tokens, and the decisions behind them — in a form AI agents can reason over, and prove it actually worked.
Action
Architected Surge Intelligence as an information layer first and skills second: a maintained knowledge base underneath six Claude skills spanning design, UI content, and engineering handoff. Built the measurement system alongside it — deterministic code checks plus expert visual review — and benchmarked v0 over 54 samples against a non-intelligence baseline.
Result
Surge imports resolve went 7% → 96% and type-checks 0% → 74%; adherence rose 0.41 → 0.63 and expected components used 42% → 68%. The eval also exposed exactly where the system still fails — layout composition, tables, empty states — turning quality from an opinion into a prioritised backlog.
Global Design System
Bumble Inc. · 2022–2026
Situation
Bumble was moving to a multi-brand, multi-product strategy on a foundation that couldn't carry it. Each brand had its own visual language, systems and implementation approach; the design system was disconnected from the codebase, with UI kits, design files and documentation maintained by hand in Figma. Tokens were managed manually through an outdated architecture, adoption across iOS, Android and Web stayed low, and the DS team was absorbed into reactive support — running manual validations before every handoff and answering the same questions about usage and standards.
Task
Rebuild the foundation so that one system could serve three brands across three platforms, and make it hold without the DS team standing behind every piece of work. That meant three things at once: a token architecture designed for brand and platform separation, a synchronised pipeline between design and code, and enough automation and documentation for product designers to move on their own.
Action
I led the re-architecture and its adoption end to end.
Foundation. Defined a semantic, multi-layer token architecture (core → semantic → component) that separated impact from reach, letting every brand and theme share one structure. Systematised the UI foundation and rebuilt Bumble's UI on top of it, then evolved components against real product scenarios rather than in isolation.Infrastructure. Re-architected the token pipeline: Tokens Studio as the source of truth, connected to GitHub and Figma, with a transformer layer syncing values into iOS, Android and Web formats and documentation regenerating from the same source. One update, full sync.Automation. Mapped the friction points across the design lifecycle and automated them — a custom Figma plugin validating token, component and pattern usage so designers get feedback before handoff instead of in review, plus ready-made patterns and templates at the design phase and automated QA checks at the end.Adoption. Grew and led a dedicated DS team by securing leadership buy-in, ran cross-team design reviews to hold system integrity, built the documentation hub that made onboarding self-service, and drove company-wide efforts — rebranding, accessibility — through the foundations rather than as separate projects.
Result
The system went from a manually maintained library to infrastructure the company builds on: design and code stay in sync from a single source of truth, and scaling to a new brand or theme became a configuration task rather than a project. A rebrand that would have taken months ran 4× faster. Design QA issues dropped 89%, coordination effort on updates fell ~70%, maintenance effort 60%, and automation gave the team back ~8 hours a week. New projects reached 100% adoption, with designers shipping system-aligned work without the DS team in the loop.
Situation
Vodafone's digital presence had grown market by market and channel by channel. The app, the web, in-store retail screens and the chatbot each carried their own interpretation of the brand, and every one of the 15+ local markets adapted it again. The result was a customer who moved between channels and didn't quite recognise the same company — and design teams across markets solving the same problems in parallel, from scratch.
Task
Define a new digital visual identity for Vodafone and make it real across the group. Not a brand deck, but a direction proven in a live product first, then translated into the systems, components and governance that let four surfaces and fifteen markets adopt it without drifting apart.
Action
I led the definition of the new UI direction and drove it through to implementation, from the core design team.
Proved it in the app first. Took the identity from principles to a shipped product surface — typography, colour, spacing, iconography and interaction resolved against the real constraints of a live app rather than in a mockup. That gave the direction something to argue from: a working product instead of a proposal.Systematised it. Turned what the app validated into foundations, components and patterns, so the identity travelled as a system others could build with rather than as a set of guidelines to interpret.Rolled it out across channels. Led the extension to web, retail and the chatbot experience — each with different technical realities and constraints — holding one coherent expression across all four while letting each surface behave the way it needs to.Scaled it across markets. Guided the product design teams to deliver system-aligned output, and partnered with engineering and product to scale usage through governance and tooling, balancing group-level standards against genuine local needs.
Result
One digital identity, consistently expressed across app, web, retail and chatbot, and adopted across 15+ markets. Consistency scores in UX audits improved 62%, and duplicated design work across markets dropped 50% — teams that had each been rebuilding the same surfaces were now building on shared foundations.