The problem
Movie discovery is stuck on genre grids. Users want mood-based discovery, transparent recommendations, and a way to connect viewing history with reflection.
My role
Solo design + engineering. Architecture, data model, AI pipeline, UI, deploy.
What it does
A Netflix + IMDB hybrid built on Apple's visionOS material language. Onboard with 10 film ratings to seed a personal taste embedding; recommendations stream live from a 2D mood dial (valence × arousal) with three transparent reasons per pick. A nightly cron auto-renders a shareable recap card from your journal entries — Wrapped-style, but personal. Shipped solo in 6 weeks on free-tier infrastructure.
Stack
- Next.js 16
- TypeScript
- Tailwind v4
- Framer Motion
- Neon Postgres
- pgvector
- Drizzle ORM
- Upstash Redis
- Clerk
- Vercel AI SDK
- Google Gemini
- Mux (HLS)
- TMDB
- Serwist (PWA)
- Vitest
- Playwright
Tags
- #AI
- #RAG
- #Next.js
- #Liquid Glass
- #pgvector
- #PWA