An Arabic-first basketball statistics platform for Syrian leagues: a public site for standings, matches, players and teams, and an admin dashboard with two AI features — box scores read from photos, and a chat assistant whose edits wait for approval.
Syrian leagues needed one Arabic place for results, standings and player statistics — and entering those stats by hand from FIBA box scores was slow and error-prone.
I designed the interface in UX Pilot, managed the project, then built it full stack as two independent apps: a Next.js public site and admin dashboard, and a NestJS API on Prisma and MySQL where standings and averages are computed on read, never stored.
The admin uploads a FIBA box score and the model extracts it under a strict JSON Schema. Zod then checks what the schema can’t — no negatives, made ≤ attempted — and player names are fuzzy-matched. Nothing saves before a human review.
A streaming (SSE) chat answers questions about the data instantly. Its write tools never change anything: they stage a pending action the admin approves from a card in the chat, which then runs through the same validated services as manual entry.
Any league, season, match or player can be marked Prime. Fans sign in separately and subscribe by month or year; the admin activates each one with a three-level scope — leagues, seasons, phases. Locked pages return a teaser instead of the data.
Standings come from a pure, unit-tested function with the client’s FIBA tiebreakers: head-to-head only counts once all its scheduled meetings are played, otherwise overall point difference decides. Every compound write runs in a transaction, which also clears the season’s cache.
Two independent apps, back end and front end, deployed automatically by GitLab CI to a VPS I manage, with PM2 keeping them running.
Result: launched in September 2026 at 963court.com, with Prime content and fan subscriptions.
