ULTSPOTA K-pop Fan Trip Planner for Visitors to Korea

A K-pop Fan Trip Planner for Visitors to Korea

ULTSPOT

Background

Birthday cafes and pop-ups are posted in Korean text and poster images, and events that run only three to five days overlap with permanent spots. Overseas fans who know where they want to go still struggle to turn it into a plan that fits their travel dates.
Fan communities showed repeated questions about English searches returning no events and whether you just buy a drink or the goods are sold or given away, and the Korea Tourism Organization itself introduces birthday cafes as a fan culture international visitors experience. The goal was to ship this as a working service within three days for the Wanted AI Championship 2026.

As PM of a 5-person team (four teammates on business planning and UI/UX), led planning, full-stack development, and deployment. Wrote the plan v0.5, PRD, functional spec, and user flow, and split features into P0/P1/P2 to fix a three-day scope. Crawling, payments, rankings, and distance-based route optimization were left out.
Rechecked public sources, acknowledged that competitors already offer English pages and course features, and narrowed the differentiator to picking events that fit travel dates and booking conditions, explaining entry conditions, and adjusting the plan.

As the team's only developer, built the 4-step planner on Next.js 16 and Supabase with Kakao directions, the Korea Tourism Organization TourAPI, and Google Maps. Split the work into 78 tasks run in parallel with AI coding agents, and shipped after E2E checks at three screen widths.
Hours, travel time, and gap math run in deterministic code while AI only ranks nearby candidates by taste. The same input gives the same plan, and unknown hours are shown as 'unconfirmed'.

Wanted AI Championship 2026: ranked 52nd of 1,390 projects with 74 likes (top 3.7%)
Shipped at ultspot.vercel.app (266 commits, 78 tasks, Sep 16–28 2026)
Linked 243 artists and 12 verified places, with a 4-language UI
Recommendation model matched 30/30 synthetic test cases (median 0.54s); UI accessibility audit scored 17/20