UK Coffee Finder

A coffee matchmaker that champions UK roasters

28 days • ~850 prompts • ~700M tokens

Why I built it

This was my first experiment using AI tools to build a working product from scratch. I worked as a barista in Sydney and I love coffee, and friends and family kept asking me where to start, so I was inspired to build something that could help them.

Process

I used Claude's deep research functionality which included me feeding in 30 Reddit threads. This picked up on themes from coffee drinkers in the UK which helped me understand what to design - it should show one result, use easily understandable wording and stay impartial.

I turned the research into a product document and a build brief for each phase, so each stage started with the correct context. I built it in Claude Code with VS Code to learn how the project was built and make manual edits myself. I made the illustrations and icons in ChatGPT, simplified them into SVGs, and designed the results screen in Figma so the build would match it.

The solution

You're asked a few simple questions about what you own, what you'd spend and how you like your coffee. Depending on your answers, the result is one method, one bag of beans and where to buy it.

The hardest part was getting the logic right. Every question had to feel necessary - enough to give an accurate recommendation, but not so many that it felt overwhelming. All 77 blends from 40 roasters were categorised by flavour, intensity and roast level and are regularly checked against the roaster's site by an agent that I set up.

What I'd do differently

Lock in the visual direction and finish the design system before building, and set up skills on day one. I built most of the experience before the look was finalised, so many screens had to be redone and it burned through a lot of tokens.

The result screen, before and after the Figma redesign