predictive sports intelligence

Three independent sports pipelines running on a single NVIDIA DGX Spark — each with its own database, model, scheduler, and public API. Pick one below, or scroll down to see how the stack is put together.

The 3 projects

Project 01

Men's College Basketball

D-I game predictions, biomech analysis, and draft scouting with the False Floor big-board model.

checking…
Open MCBB dashboard
Project 02

Women's College Basketball

Full WCBB pipeline — daily predictions, prospect tracking, and draft projections from the Spark stack.

checking…
Open WCBB dashboard
Project 03

College Football

CFB game predictions and draft scouting, powered by 18 scrapers feeding the Spark pipeline.

checking…
Open CFB dashboard

How it's built

Tier 1 · Public web

Cloudflare Pages + Workers

  • Landing: sportai-umbrella.pages.dev
  • 3 Worker APIs — one per sport
  • Handles CORS, caches prospect lists 5 min
  • Never sees the Spark directly
Tier 2 · Private tunnel

Cloudflare Tunnel

  • sport-tunnel.eduleaderai.org
  • wcbb-tunnel.eduleaderai.org
  • cfb-tunnel.eduleaderai.org
  • Outbound-only · no public IP
Tier 3 · Compute

NVIDIA DGX Spark

  • 3 FastAPI services (ports 8888 / 8001 / 8002)
  • 3 Supabase stacks, isolated per sport
  • n8n · Firecrawl · 37 systemd timers
  • Models trained locally on GPU

Per-sport config

Sport FastAPI port Supabase ports Tunnel hostname Public worker
● MCBB 8888 54322–54327 sport-tunnel.eduleaderai.org sport-real-api.eduleaderai.workers.dev
● WCBB 8001 54344–54348 wcbb-tunnel.eduleaderai.org wcbb-real-api.eduleaderai.workers.dev
● CFB 8002 54334–54338 cfb-tunnel.eduleaderai.org cfb-real-api.eduleaderai.workers.dev

Runtime sandbox

Primary compute
DGX Spark
192.168.0.56 · GPU + CPU
Secondary / fallback
Mac Mini
warm backup · daily health probe
Archive / storage
UNAS (16TB HDD + 2TB SSD)
192.168.0.251 · RAID5
Edge
Cloudflare
Pages · Workers · Tunnel · DNS