Vyne

Problem

Brand–influencer matchmaking today happens over DMs and spreadsheets: a brand cold-messages a creator (or vice versa), and there's no structured signal of mutual interest before both sides invest time in a conversation. Budget and rate expectations usually surface only after several messages, by which point the mismatch has already cost both sides time. Vyne's bet was that a lightweight, mutual-consent discovery layer — swipe, match, then talk — could surface fit (niche, budget, follower range) before either side commits to a conversation.

Approach

Rather than bolt a swipe UI onto a single generic 'user' model, the data model treats businesses and influencers as distinct profile types (separate tables, separate signup forms, separate discovery filters) unified by one matching engine: a `matches` row tracks independent `business_liked` / `influencer_liked` flags, and a match is only confirmed when both are true. Swiping is one path into that engine; a second 'Matches' surface (Likes You / You Liked, with Approve/Decline) exists for people who'd rather browse than swipe blind — same underlying state machine, two entry points. Auth and account lifecycle got the same rigor as the core loop: deep-link refresh (landing on /Matches directly must not bounce to a 404), back-button safety after login, stale-session recovery after long idle tabs, a full forgot/reset-password flow, and a deliberately friction-full 3-step account-deletion flow (re-verify password → type to confirm → server-side deletion via a Supabase Edge Function) so an irreversible action isn't one misclick away.

Solution

Shipped an end-to-end product — 13 screens from landing to account deletion — running on Supabase (Postgres, Auth, Storage, Edge Functions) and deployed on Vercel, with messaging refreshing on both sides within ~2 seconds and a documented end-to-end test checklist covering signup, matching, messaging, and every account-recovery edge case.