AI Platform
Zivvi
Discovery platform and ranking engine for 2,300+ web apps across 28 categories, combining daily Tranco global traffic velocity, hybrid Upstash Vector + Postgres search, and Next.js 16 Partial Prerendering.
- Role
- Founder & Engineer
- Live site
- zivvi.app
Highlights
- Daily ingestion and 30-day momentum scoring against the Tranco top-1M domain dataset (CrUX, Cloudflare Radar, Umbrella, Majestic).
- Hybrid search fusing Upstash Vector semantic embeddings (text-embedding-3-small) with Postgres full-text and pg_trgm typo-tolerant search.
- Next.js 16 Cache Components and Partial Prerendering (PPR) with shared static app shells for instant navigation across thousands of app, category, and alternatives pages.
- Verified maker claim flows, personal app hubs, embeddable SVG badges (/badge/[slug]), public JSON API (/api/v1/apps), and /llms.txt.
The problem
Most software directories either spike for 24 hours on launch day and disappear, or rank tools by whoever buys the biggest ad package—leaving indie makers without compounding discovery and users with stale lists.
What I built
A web app marketplace at zivvi.app that separates objective global traffic rankings (/popular) and 30-day rank velocity (/trending) from verified community upvotes and reviews (/top), alongside programmatic alternatives and built-with stack pages.
Search, data & performance
Scheduled cron pipelines sync daily Tranco domain ranks, GitHub repository stats, and vector embeddings across 2,300+ apps, served through Next.js 16 Partial Prerendering and segmented XML sitemaps.
Stack
- Next.js 16
- React 19
- TypeScript
- Supabase
- Upstash Vector
- Tailwind CSS
- Tranco
- Vercel