Manas Dalvi

All projects

06ML · MLOps

Premier League Prediction

End-to-end MLOps platform forecasting match outcomes with multi-source data fusion and explainable AI.

View source on GitHub

Stack

Prefect · PostgreSQL · XGBoost · MLflow · SHAP · FastAPI · Streamlit · GCP

In one line

zero-leakage feature store

Overview

An end-to-end MLOps platform that ingests football data from multiple sources, engineers predictive features, and serves explainable match-outcome forecasts on automated weekly pipelines with full experiment tracking.

Architecture

Demo

Demo coming soon

Highlights5

01

Automated Prefect pipeline ingests weekly data from the FPL API, Understat, and football-data.org into a time-indexed PostgreSQL feature store with zero data leakage.

02

Engineered features include ICT-weighted squad-depletion scores, rolling xG/xGA averages, and xG-overperformance regression signals.

03

Fine-tuned twitter-roberta-base-sentiment for pre-match NLP sentiment on manager press conferences.

04

XGBoost models tracked with MLflow, evaluated with Ranked Probability Score and full SHAP-based per-prediction explainability.

05

FastAPI inference API on Google Cloud Run with a Streamlit frontend for what-if injury simulation and Kelly Criterion backtesting.

Outcomes

Zero-leakage
time-indexed feature store
RPS + SHAP
eval + per-prediction explainability
3 sources
FPL, Understat, football-data
NextPRism