Work

One Design System for Multi-Product Platforms

Built Nourish's first true design system from scratch across three competing front-end stacks, drawing on the same discipline I used to build Nightingale at Florence.

Organisation
Nourish Care, Florence
Project period
2022 — Present
Link
Disciplines
  • Design systems
  • Governance
  • Accessibility
  • AI-assisted workflows

In brief

The problem
A new, small design trio in a business only just moving to agile, with only legacy kits to rely on, while the platform added new products fast across three competing front-end stacks.
My role
I led it end to end, from the case and executive buy-in through audits, rollout and foundations, and designed many of the components myself.
What changed
One system each product implements natively, adopted inside existing roadmaps. Once it was solid enough to build on, AI-assisted builds became viable, and the conversation moved to automating component builds, documentation and tokens, and to which parts of our flow agents can take on.
Outcome
733,000+ insertions in 12 months, under 0.5% detach and 10–14% of design capacity returned. AI-assisted components land at 70–80% first-pass quality, so designers' time shifts from pixel-pushing to review and judgement.

Context

When I joined Nourish, design was a small team of three, myself included. Until then it had worked as a service to the rest of the business, taking requests ticket by ticket, so there had been little time to invest in shared foundations, and screens were built from legacy UI kits. The company was only just moving to agile under a new CPO, and the platform was growing fast, with several new products due for release over the next few quarters. It was my priority to expand the team, put design processes and ops in place, and lay foundations robust enough to scale with the platform, all at once.

The case I made for a design system was a strategic one rather than an aesthetic one: build a higher-quality product, faster and for less, in a way that holds up as the team and platform keep growing.

I'd been here before. At Florence I led and built Nightingale (opens in a new tab), a cross-platform design system with its own governance model (a health-status tracker for component maturity, and a decision matrix to prevent input components being misused). I then hired a junior designer to take on ownership of it and championed them, staying in the weeds as a partner on components and decisions - sometimes building, sometimes developing parts like the type system, sometimes reviewing. Pulse carried that same instinct further, this time with the usage data to back it up.

My role

I led the programme end to end - building the strategic case and winning executive buy-in, the design audits, the rollout strategy and the foundations - and designed many of the components myself, from buttons, tags and cards through to side sheets, the core inputs, selection controls and navigation. I worked closely with design and engineering leads across all three front-end stacks throughout.

Constraints

Most design system stories assume one codebase, whereas Nourish had three, each with its own patterns, component logic and technical constraints. A component that's simple to define in Figma could mean three quite different engineering problems depending on which stack it lands in.

That forced a choice for every component: reskin the existing UI for a quick, visible win, or rebuild it properly. There was no one-size answer, so we worked it through case by case, closely, with the teams and products involved.

What we did

Discovery

We started with a design audit across core flows, to understand where UI was genuinely custom and where components could be consolidated.

Decisions

Rather than forcing a big-bang migration, we set one operating principle for legacy areas: if a team is already touching a page for other work, that's when it moves to Pulse. Migration happens as a by-product of work already planned, so it never has to compete with feature roadmaps.

We also chose to roll out by flow rather than by feature. Going feature by feature would have phased out legacy styling faster, but risked a patchwork UI where one journey feels half-migrated. We anchored early effort on the highest-impact, most visible flows, so the earliest wins were also the ones people noticed - which mattered for keeping teams on board before they'd felt the benefit themselves.

Accessibility was built into the foundations from the outset rather than retrofitted. Colour, typography and component states are defined against WCAG 2.2, with contrast and labelling suited to care workers in time-pressured environments.

Governance

Each page in the library carries its status (pending review, not signed off, deprecated, or due a redesign), so designers can see at a glance whether a component is ready to use.

List of design system pages, several marked pending review, not signed off or deprecated (opens full size in a new tab)
Pulse pages, each labelled with its status.

Holding up on a complex screen

The MAR chart has since been rebuilt on Pulse. It's one of the densest, most safety-critical screens in the product, so it was a good test of whether the system holds up under real pressure.

The medication chart rebuilt with Pulse components, showing three fictional people and their medications (opens full size in a new tab)
The MAR chart rebuilt on Pulse.

Outcome

Pulse now gives Nourish a strong framework that delivers consistently across every product on the platform, with each one free to implement natively rather than being forced into a shared codebase. Adoption is happening steadily inside existing team roadmaps, rather than stalling as a separate project competing for time.

Pulse's maturity is also what made AI-assisted builds viable in the first place. Once Foundations and Components were solid enough to build on, components assembled with Claude and Figma MCP were landing at 70–80% first-pass quality, against a rough 30% for the same build done manually in the same time. For me, that says more about the strength of the design system than about the AI tooling. There's more on how the team works with AI on the Leadership page.

Reflection

What carried over from Florence wasn't a particular pattern library so much as the operating model - shared ownership between design and engineering, incremental rollout over big-bang rewrites, and documentation treated as seriously as the components themselves.

Code Connect coverage sat at around 7% when we started pushing on AI workflows, a foundational gap I hadn't prioritised early enough. It's now a standing step for every new component.

Storybook has also needed real upkeep as our documentation source of truth. As workflows speed up, especially AI-assisted ones, documentation debt builds faster than you'd expect, so it needs its own audit cadence.

From this project

The detail behind it