
About the project
In January 2026 I found Riftbound, Riot Games' trading card game. After years of League of Legends, I was hooked from the first match. The more I played, the more the tools for players annoyed me: unfinished, hard to read or locked behind a paywall.
In April I did the only thing I know how to do with frustration as a designer: I opened Figma.
That's how RiftBuddy started: a card browser, deck builder, collection with a card scanner and tournament meta, all in one place. I designed and shipped it on my own, from idea to production, and in the three months after launch I delivered five more versions. The home page says "Built by a player for players". It's not a slogan. That's just how it went.
role
Product designer, product owner, sole creator
platform
Web, mobile first, installable on the iPhone home screen
stack
React 19, TypeScript, Tailwind, Supabase, Sanity, Vercel
timeline
April 2026 → public launch in June, before the Regional Qualifier in Utrecht. Still live and in active development.
Problem
Before a tournament I was jumping between five tabs: Riot's card gallery, a collection spreadsheet, a separate deck builder, results sites and articles with decklists. None of them connected what's in my box with what's winning tournaments.
Riftbound is a young game, so its tool ecosystem was only just forming. I saw a window: whoever gives players one place first builds the habit.
Who I designed for
- Collectors want to know what they're missing for a full set, without typing cards in by hand.
- Casual players want to build a deck from what they own and show it off.
- Tournament players want to see what wins and quickly copy a decklist from a Regional Qualifier.
I decided early that the tone and features had to work for all three groups, not just the competitive scene. Hence guest mode with no account required, and collection badges with no leaderboards or streaks.

Process overview
Click any of the steps to swoosh to the section
How I worked
Design first, code second. Always. Before a single line of code existed, I had a design system in Figma (navy #0A0E1F, yellow #FFC400, cream #F0E6D3, Satoshi and Nunito, a 4 px grid), seven user flows, a screen inventory with separate mobile and desktop layouts, and a competitor analysis. I didn't only look at Riftbound tools, but also at how Lorcana, Magic and Flesh and Blood solved the same problems in their apps. The logo is two fists in a fist bump, yellow and cream: two players on opposite sides of the table, bumping fists before the match.
AI as my engineering team
I'm a product designer, not a developer. I built RiftBuddy alone by treating AI (Claude Code) as an engineering team that I manage. I decided what gets built and why, how it looks and whether it works in a player's hands. The AI wrote the code to match those decisions.
I also learned one thing fast: don't trust summaries. The AI could report with full confidence that it had fixed something in two files, and a search through the repo would turn up a third. So I kept a running log of process mistakes, like "a fix must cover every instance of an element, not just the one I pointed at". Every entry is one real slip-up that never happened again.
One phase, start to finish
To keep it from falling apart, I worked in small, closed phases:
- Bug report or player idea
- Product decision and scope
- Phase plan with success criteria
- Implementation and tests
- Test on iPhone, on a Vercel preview (back to fixes if needed)
- Production
Every change went to a test branch and a Vercel preview first. I checked it on my own phone, launched from the home screen icon, the way players actually use the app. Only what passed that test went to production.
Key features
Six areas, each answering one player need.
Card scanner
Point your phone at a card and it lands in your collection. Full-screen camera, slide-up confirmation panel, an "Added" bar with Undo and a session counter.
Collection & badges
See how much of each set you own and what's missing. 46 badges for milestones (from 50 to 1180 cards, full sets, domains), synced across devices.
Deck builder
Build a deck with filters by domain and keyword, with live legality checks. Import from 4 formats (text, CSV, TTS, deck code), with a report of lines that didn't match.
Deck export to PNG
One click and the deck graphic is ready for Discord or Instagram. Template designed in Figma, 1920×1048 px, rendered in the browser.
Meta & tournaments
Tier list, results from TopDeck.gg and decklists from 6 Regional Qualifiers. One shared deck view across all pages: average cost, energy curve, placement colour.
News
Articles about the game and the app, written in Sanity Studio and published without touching code.
And more
Plus a browser for 1180 cards from 5 sets, a glossary of mechanics, guest mode, and sign-in with an email code that also works in the app installed on an iPhone.












Challenges & decisions
The most interesting part of this project wasn't the screens. It was the decisions: what to build, what not to build, and what to trust: the data, the code or the phone.
Deadline: Utrecht
The app had to be public before the Regional Qualifier in Utrecht on June 12, 2026. A tournament doesn't move because I'm running late. I froze the scope: sign-in, hosting and working card images, no new features until tournament day. Everything "nice to have" went on a later list. The launch worked, and that list turned into the next versions.
Second deadline: Vendetta
At the end of July the fourth set, Vendetta, was coming out. I had just come back from three weeks away from the project, and the Pre-Rift prerelease events were days away. The database grew from 950 to 1180 cards. Along the way it turned out the list of sets was hardcoded in three places in the code, not two, as the AI's summary claimed. I made it live on the day Pre-Rift started. Rulings, bans and errata I gathered by hand from Riot's announcements and the FAQ, because no automatic source existed yet.
A human beats the scraper
A scraper pulled the tournament decklists until one run wrecked the database: wrong placements, truncated decks. I could have patched it. Instead I made the official articles the single source of truth and re-entered 6 tournaments with import scripts. Slower, but a player never sees a false result. For the same reason I don't pull data from competing sites, even when it's technically possible. In a community this small, trust is worth more than a few extra decklists.
Scanner: no new wheel
The scanner is the biggest feature in the project. Instead of building recognition from scratch, I adapted the open-source riftbound-scanner (MIT). Before releasing anything, I wrote to its author, though the licence didn't require it. He replied the same day: "Of course you can use it". One of the nicest messages I got during the whole project.
I added two product decisions on top:
- A card goes into the collection only after you tap Save. The scanner suggests, the player decides. A scanner mistake never quietly corrupts a collection.
- The scanner says when it's not sure. I set the confidence thresholds by measuring 40,000 pairs of different cards, not by eye. When in doubt, it shows 3 candidates instead of guessing.
Then the author showed me something that took the wind out of me a bit: his scanner recognises 50 cards at once. Mine scans one at a time. Same foundation, completely different scale. I'm glad he showed me, because now I know where the bar is.
Reversing my own call
At launch the collection lived only in the browser. On purpose, to make it in time for Utrecht. Then I started designing badges and saw the problem: a badge that disappears when you switch phones motivates no one. I moved the collection to the cloud and designed an explicit merge screen for when a player has different data on two devices.
Badges, no tricks
Badges reward collecting. They don't keep people in the app by force. No streaks, no leaderboards, no time-limited rewards: 46 badges for real milestones that stay forever.
The phone beats the code
Email link sign-in worked everywhere except the app installed on the iPhone home screen: the link opened Safari and the player ended up outside the app. I replaced it with a 6-digit code typed into the app itself. Same with the badge icons: the Figma export looked fine in every desktop browser, and on the iPhone half the graphic disappeared. The only test that settles anything is the player's real device.
Numbers & timeline
In under 3 months since public launch I shipped 5 versions across 24 phases. Alone. Numbers from the repository, as of August 26, 2026.
In numbers
production versions
5 (v1.0 → v1.4)
work phases
24, 22 closed
commits
744
lines of app code
≈ 22k (TypeScript/React)
screens
29
cards in the database
1180, 5 sets
badges
46
automated tests
229
registered users
40
Versions
v1.0 launch · June
Public beta before Utrecht, 6 RQ decklists, guest mode
v1.1 png export · June
Deck graphic in one click, Figma template
v1.2 seo · July
Correct domain and page descriptions for Google
v1.3 vendetta · late July
New set: keywords, domain filters, rulings and errata
v1.4 collection · August
Scanner, cloud collection, badges, iOS sign-in, Sanity Studio
The number I don't like
744 commits, 29 screens, 229 tests. And 40 registered accounts.
I won't pretend that doesn't sting. But the number says something important: a good product doesn't sell itself. Riftbound is a niche inside a niche. Competing tools have teams and booths at events. I had an X account and evenings after work. Most of my energy went into building, and not enough into getting it in front of people.
Two honest notes. First, 40 is only accounts. Guest mode works without signing up, so I don't know exactly how many people really use the app. Second, I now know where the players are: community Discords, local game stores, content creators. That's where I should have been from day one, not after version five.
What I learned
A designer who ships the product themselves sees things you can't see in Figma.
Lessons
- Freezing scope is a design decision. Utrecht forced the question "what is really needed on day 1", and that's the only reason the app shipped at all.
- It's fine to change your mind. A local-only collection was right for launch and wrong for badges. Changing a decision when new facts arrive isn't failure.
- Automation suggests, a human approves. The scanner and deck import never change a player's data without their tap.
- The real device decides. Tests, types and a desktop browser let through bugs that took a second to spot on an iPhone.
- AI is a team you have to manage. Clear scope, success criteria and a log of recurring mistakes beat long prompts. And check, don't take its word for it.
- Building the product is half the job. The other half is making sure people hear about it.
What's next: Riot is launching an official Riftbound portal, so the next decision is whether to integrate with it or deliberately go alongside it. Also on the list: scanning many cards at once, automatic card database updates with every new set, and showing up where the players actually are.