HyperTFT

Data-driven Teamfight Tactics analytics — patch-by-patch tier lists, a comp builder with live suggestions, and item recommendations, computed from ranked matches collected across six regions (NA, EUW, KR, SEA).

Live: https://hypertft.mkgg.dev

How it works

Riot API ──► collector ──► PostgreSQL ──► aggregator ──► API ──► frontend
             (slims at     match_slim     (incremental    FastAPI   Astro/React
              ingest)      + ledger        fold + deep    + Redis   on Cloudflare
                                           pass)          cache     Pages

• Collector pulls ranked matches daily (rotating region groups), extracts a compact per-match record at ingest, and discards the raw ~20 KB payload — storage stays flat as volume grows.
• Aggregator runs in two tiers: a cheap incremental fold of new matches into running placement stats every day, and a periodic deep pass that recomputes comp similarity (Jaccard), unit swap/addition deltas, and per-unit item stats over a rolling patch window.
• API serves pre-aggregated stats with GIN-indexed array queries for comp lookups and a Redis cache-aside layer; old patches stay browsable forever.

Stack

Python · FastAPI · PostgreSQL (Supabase) · Redis · Astro + React + Tailwind · GitHub Actions · Cloudflare Pages

Repository layout

Path	What
backend/	Collector, aggregator, API — setup, schema, and pipeline docs
frontend/	Astro/React site (tier list, comp builder)
backend/deploy/gcp/	Optional self-hosted deployment kit
.github/workflows/	Daily data-pipeline cron

See backend/README.md for full setup and architecture details.