KAIZEN
Blue-grey mountain peaks rising through a sea of clouds

KAIZEN

Change for the better.

Move your pointer or touch the landscape to reveal ASCII. Scroll down to dissolve the scene into code and explore. Scroll back up to return. Press P to pause or resume the clouds.

Prithviraj Patil working at his laptop

Prithviraj, also known as Kaizen

Prithviraj Patil.

Artificial Intelligence undergraduate India

I build machine learning applications, study how models learn, and use data to ask better questions about football.

Currently studying at Mahindra University, École Centrale School of Engineering.

Research & applied AI

kAIzen AI Lab

Scientific papers, learning dynamics, and geospatial models.

01 Scientific document systems

ScholAR

An answer is only useful if you can inspect its evidence.

A local research assistant that links answers to the supporting paragraphs, tables, and figures in scientific papers.

Python · Retrieval · Local language models

02 Learning dynamics · In progress

KAD

When does a strong signal slow the learning of a weaker one?

Paired experiments in recurrent networks, finite-width analysis, and mitigation studies, with reproducible experimental protocols.

Recurrent networks · Causal comparisons · Research manuscript

03 Learning dynamics · Research

Gradient Starvation

Weak learning failure alone is not gradient starvation.

Matched both-feature and weak-only experiments distinguish poor learning from causal starvation, combining finite-width response dynamics with reproducible evaluations.

Recurrent networks · Matched interventions · Finite-width dynamics

04 Geospatial machine learning

TerraMind

From satellite imagery to an inspectable plume mask.

A methane-plume segmentation prototype, developed as MethanePulse during a hackathon. Includes a dashboard, local inference API, and georeferenced exports.

TerraTorch · Sentinel-2 · GeoTIFF

Other AI experiments

Football analytics

Kaizen
Football Lab

Football questions, tested against the data.

Open the analysis collection
  1. Rondo Lab

    A reinforcement-learning playground inspired by Barcelona’s 6v2 rondo, built to revisit sixth-semester RL through football. A 2D simulator models player movement, passing, interceptions, and defensive pressing.

    Possession and defending teams take turns training with PPO against a frozen opponent. Short runs still produce repetitive passing and stalling; simulation mechanics and reward design strongly shape what the agents learn.

    The browser gallery lets you watch recorded simulations and compare real training checkpoints. Training runs locally; this is a learning experiment, with no claim of strong football yet.

  2. The Invisible Runner

    Measuring space created by players without touching the ball.

    Watch the tracking replay
  3. Rest Defence Analyzer

    How vulnerable is a team to the counterattack while it attacks?

    Inspect rest defence
  4. Pass Decision Lab

    Reviewing the visible ground-pass alternatives around the ball.

    Compare passing options
  5. Is possession actually dangerous?

    Comparing Barcelona’s possession threat with its rivals.

    Explore possession danger
  6. TacticAI Lite

    Which corner-kick patterns are associated with shot creation?

    Study the corners
  7. Possession Lab

    Who has possession when the ball disappears from the tracking?

    Open the observability explorer
  8. The Defender Nobody Notices

    Finding defenders who obstruct passing lanes before a pass is attempted.

    Inspect lane obstruction
Models & tournament stories 5 projects

Expected goals, from scratch

Shot geometry, context, and calibration on StatsBomb open data.

World Cup predictor

Match probabilities and tournament simulations using Elo, Dixon-Coles, and XGBoost.

World Cup stories

A React dashboard exploring tournament goals, squads, clubs, and shootout history.

Public event and tracking datasets, alongside simulated training drills. Each project links to its methods and limitations.

Education

Undergraduate

Artificial Intelligence

Mahindra University

École Centrale School of Engineering · India

Elsewhere on the internet

For a question about the work, a research idea, or a conversation about football:

prithvisp28@gmail.com