Building a product from scratch at an early-stage startup
Eleven months at a remote AI startup, first designer in. From no brand or UI to a shipped product with a successful pilot demo.
- Role
- Senior Product Designer & Design Engineer
- Client
- lumenx
- Date
- Jun 2025 – Apr 2026 (11 months)
- Team
- Founder, CTO, principal scientist, and frontend developer. Fully remote across UK, France, USA, Canada.
- Scope
- Brand, marketing website, product UI, design system, production code contributions
- Tools
Claude AI
Figma
GitHub
Linear
Shadcn
Adobe Illustrator CC
Notion
Tailwind CSS
A note before diving in: since this work was completed, lumenx has continued to evolve. The brand and product direction have moved on from what's shown here, as early-stage startups often do.
Context
An AI behavioural research startup with everything user-facing still to build.
lumenx had built the backend. Healthcare and risk & compliance were the target verticals. The product was structured around established behavioural science frameworks. Small team, fully remote. I joined as the first designer.
The work
Designing the project setup, manual reverse-engineered from the AI.
The product had three surfaces: setup, AI interviewer, and reporting. The deepest design work was on setup.
Two routes were live: manual setup for owners who knew what they wanted, and an AI-led intake that asked the right questions and inferred the project shape from the answers. The AI intake started as a vibe-coded experiment from the founder, which I finessed and shipped. The manual setup came after, reverse-engineered from the AI flow, mapping the data the AI captured into a structured field-based form.
The principal scientist was a strong collaborator on the data architecture: which data was AI-generated, which was user-entered, and which still needed AI intervention regardless of route.
Working in code
The codebase as source of truth.
Early in the engagement I pivoted the foundation from custom TypeScript and SCSS onto shadcn, Radix, and Tailwind. Any developer could then pick it up faster.
From there I contributed components and pages in production code directly, raising PRs alongside the frontend developer. .md instruction files and MCPs kept AI-generated output inside the system.
The 5-day pre-demo refactor
Foundations before polish, even with no buffer.
Five working days before the pilot demo, a new report section had been built using AI tooling. The code had bypassed our Tailwind and shadcn foundations and was written in raw CSS. It worked but was visually inconsistent.
With the frontend developer on holiday, I refactored. Migrated the raw CSS back into Tailwind and shadcn, restored the core components, isolated demo-specific changes into versioned component copies.
The time cost came up mid-process. I explained that polishing the UI on top of broken foundations would be like driving a car with a flat tyre. The work continued. Demo shipped on time. Pilot landed.
Outcome
A shipped product, an inheritable system.
A live website. A working app for creating projects, conducting behavioural interviews, and generating structured reports. A pilot demo that landed with a key enterprise prospect. A design system on shadcn/ui, Lucide, Tailwind, and Radix UI, inheritable by any developer.
What I took from it
Owning the full stack taught me how the layers fit.
Brand influences product, product constrains system, system shapes what's possible next sprint. You don't see those connections until you've held them all at once.
Working in code as a designer changes how you design. Guardrails on AI tooling keep the output inside the design system by construction, not by review.
And learning a new domain in behavioural science alongside the build was the part I took the most from.
