probODds Platform

A multi-sport prediction platform: gradient-boosted ensembles for basketball player props and calibrated specialist models for soccer, served via FastAPI with a Next.js operations dashboard, all running as hardened Docker services with on-box CI/CD.

Stack

Engineering highlights

• Ensemble models (XGBoost/LightGBM/CatBoost) with Optuna tuning and automated daily/weekly/monthly retraining pipelines.
• 8-container Docker Compose stack (2 APIs, orchestrator, bot, collectors, Postgres, PgBouncer) with health checks, resource limits, and no-new-privileges hardening.
• Self-hosted GitHub Actions runners doing lint → test → multi-stage build → blue-green deploy on every push to main.
• nginx reverse proxy with Let's Encrypt TLS, immutable asset caching, and security headers; nightly pg_dump backups with 30-day retention.

Shape

Scrapers → Reader (LLM) → Quant (ensemble) → Risk → API → Next.js dashboard, over Postgres + PgBouncer