back

Prem Performance

Prem Performance preview

Python · Dagster · Polars · DuckDB · scikit-learn · FastAPI · Next.js

Premier League performance engine ingesting 5 seasons from 3 sources into a unified DuckDB warehouse via an idempotent Dagster pipeline. Improved match prediction accuracy from 40.9% to 44.6% over baseline with rolling form and xG features.

football-data.co.uk ──┐                                        ┌── League Standings
(CSV, 20 seasons)     │                                        │
                      ├── Dagster ──► Parquet ──► Polars ──►  DuckDB  ──► FastAPI ──► Next.js
FPL API ──────────────┤   Assets      (raw)      (transform)  (warehouse)   (API)     Dashboard
(JSON, live)          │                    │                       │
                      │                    │                       ├── Team Detail + Form
Understat ────────────┘                    ▼                       ├── Head to Head
(xG, 12 seasons:                  Azure Blob Storage                ├── Model Performance
 2014/15-2025/26)                 (data lake backup)                └── Predictions vs Actuals
Ask me about Harris
Ask me about HarrisBeta
Hi! Ask me anything about Harris's projects, skills, or experience.
Info may be outdated