Yourt is a festival app that syncs the lineup, your friends and the shared weekend budget. It started as a personal challenge — solve the one question that keeps coming up on every festival, and build the whole thing with AI in the loop.
Aligning which sets you and your friends want to attend used to be a tedious job: screenshots of the timetable, a group chat, and someone always ending up at the wrong stage. I wanted to fix that — and to find out how far AI-assisted development actually gets you on a real, multi-user product rather than a demo.
A team of one, with agents doing most of the typing. That shifts the job rather than shrinking it: less time writing code, far more time deciding what should exist, reviewing what came back, and keeping the architecture coherent.
Use the left and right arrow keys to scroll through the screenshots.
Every act on the timetable, by stage and by day. Mark the sets you want to see and the app immediately shows who else in your crew marked the same one — the negotiation that used to happen over screenshots now happens inside the schedule itself. Clashes are visible while there is still time to decide, not once you are standing in front of the wrong stage.
Stages, food courts and facilities on a map of the actual site, with your friends’ live positions on top. Shared pins and status notes carry the rest — “meeting at the flag”, “we moved to Mainstage” — so a lost crew is a glance rather than a phone call that will not connect on festival reception.
A full social layer: friend requests, crews and groups, and QR code invites so joining a group at the gate takes one scan instead of a username somebody has to spell out over the bass.
Location, marked sets and lists are shared per person and per group, so you decide what each circle sees. Sharing your position with your crew for an afternoon should not mean broadcasting it to everyone you ever added.
Who paid for the camping gear, who owes what, and one balance per person at the end. Group totals, individual shares and a settle-up flow, so the weekend does not end in mental accounting.
The supply run as one list the whole group edits — what is still needed, who added it, who bought it, with progress across the crew. Nobody comes back with the third pack of the same thing.
Rate every set on music, energy and show, drop a comment, and settle the only argument that matters on the way home: which act actually won the weekend.
React Native and Expo put the same app on iOS, Android and the web, with Supabase behind it for auth, data, realtime sync and row-level security.
What follows is the short version: the three-week arc and the first learnings. The full write-up — every challenge, every dead end, and what AI was genuinely good and genuinely bad at — will replace this section.
Data model, auth and the festival/lineup core. The first week decided how far the rest could go: get festivals, sets and marks wrong and every social feature after it inherits the mistake.
Friends, groups, QR invites, privacy rules and live locations — the part where a single-player planner turns into something worth opening with other people.
Expenses, shopping lists and set ratings, then the pass that makes it feel like an app: navigation, empty states, and the same build running on iOS, Android and the web.
When implementation stops being the constraint, deciding what not to build becomes the whole job. The three weeks were spent on judgement far more than on syntax.
Agents happily produce a schema that works for the screen in front of them. Privacy rules, group membership and shared expenses need one that works for every screen after it — that part stayed mine.
The quality of the product tracked the quality of my review loop, not the quality of the prompts. Reading every diff carefully was the actual work.