← All work

BookMatch

Swipe-based book recommendations. A Next.js swipe interface on an async FastAPI + PostgreSQL API, with pgvector matching readers to books.

Problem

Picking the next book is slow when the options are a long list and a star rating. A swipe gives a fast yes/no signal, and every swipe is data the recommendations can use.
Built by a team of two, so the API contract, migrations and deployment had to be clear enough for both of us to change them safely.

Next.js and TypeScript frontend with Tailwind: sign-in, the swipe deck, and a library of saved books with sorting and filters. The UI is available in 5 languages.
Async FastAPI backend on PostgreSQL (Supabase) with JWT authentication; profile pictures and book covers stored in Cloudflare R2.
Recommendation engine: books and readers live in one shared vector space, and pgvector KNN search finds the books nearest to each reader.
Schema managed through 20+ Alembic migrations, with row-level security on the database.

Recommendations inside PostgreSQL with pgvector rather than a separate vector store, so matching, users and swipes share one database and one set of permissions.
Nightly backups from a GitHub Actions workflow: pg_dump to Cloudflare R2 with 14-day retention, because a hosted database with real accounts needs a restore path that doesn't depend on seed scripts.