TripMatch
Find classmates whose travel plans almost match yours — then turn them into one trip everyone accepts.
🏆 Grand Champion (1st Place) — 7th Annual AI Global Business Analytics Case Competition, USC Marshall School of Business (October 2026)
Live app (USC sign-in) · 🎬 Demo video · 📊 Pitch deck · Developer guide
The problem
About 500 USC Marshall MSBA students share the same academic calendar, yet trips keep falling apart. In our survey of 47 MSBA students (Oct 2026):
- 74% skipped a trip in the past year because they had no one to go with
- The top reasons were coordination, not a lack of people: no car (49%), didn't know anyone interested (49%), dates didn't match (43%)
- 72% wanted a tool to find similar plans, but 81% worried about traveling with the wrong people
The gap isn't people. It's visibility into whose plans almost overlap.
What TripMatch does
| |
|---|
| Post a plan | Dates (a break or a few weekends), places, budget, vibe, group size. Up to 5 plans per student. |
| Match | Pair and group MSBA students , ranked by a 0–100 score, each with the reasons it fits. Anonymous until everyone is interested. |
| Chat | When everyone taps Interested, a group chat opens — talk before anyone commits. |
| Lock the trip | One member sends the final plan (dates, place, cost split); everyone accepts. Contacts and a calendar file unlock. |
| Discover | Confirmed trips can post open seats. Classmates without a plan can join instantly — and leave any time. |
How it was built
- Discovery & research — user segmentation, a 47-response survey, and a product hypothesis.
- Prototype — first version in Lovable (AI no-code builder).
- Production app — rebuilt with Claude Code (AI coding agent) for custom matching logic, real USC sign-in and database-level privacy.
- Voice of the customer — 10 classmates tested the live prototype. Three changes shipped from their feedback:
- "Only if it's a small group" → group-size preference (2 / 3–4 / 5–6 / any), Strict or Flexible
- "I'd want to talk first" → chat opens at a mutual match, before anyone commits
- "What if I change my mind?" → leave any trip; the seat reopens on Discover
Technical highlights
- Explainable matching engine — hard rules (shared days, a place everyone accepts, overlapping budgets, strict group size) then a weighted score: dates 45 · destination 20 · group size 15 · budget 10 · style 10. Every score comes with its reasons. Groups of 3–4 require every pair inside to fit too. Details: docs/images/match-score.png.
- AI itinerary with guardrails — Claude drafts the shared itinerary from the options the rules engine has already proved feasible, and every AI suggestion is re-checked against everyone's dates, budget and destination before anyone sees it (
src/lib/ai-merge.ts).
- Privacy by design — Postgres Row Level Security on every table; classmates see plans, never emails or private notes; contacts unlock only inside a confirmed trip; trip changes go through locked-down database functions (no double-booking, no race conditions).
- Global places — pick from popular trips or search any city worldwide; "nearby" is computed by real distance.
- Tested — 108 automated tests, including database tests that run the real schema in an in-memory Postgres.
Stack: Next.js 16 · React 19 · TypeScript · Tailwind CSS 4 · Supabase (Postgres, Auth, RLS, Storage) · Anthropic Claude API · Vercel · Vitest
Phone / laptop ──HTTPS──▶ Next.js on Vercel ──▶ Supabase (Postgres + Auth)
│ pages, server actions │ Row Level Security on every table
│ matching + merge │ trip rules in database functions
└ Claude (itinerary) ◀───┘ rules re-check every AI draft
Team — The Seven Seas (Team A2)
USC Marshall MS in Business Analytics
Run it yourself
npm install
cp .env.example .env.local # add your Supabase keys
npm run dev # http://localhost:3000
npm test # 108 tests, no keys needed
Full setup (Supabase, email sign-in, deployment): docs/SETUP_CHECKLIST.md. Architecture, matching rules and changelog: docs/DEVELOPMENT.md.
