machine learning apps
Gradient Descent Dungeon
An interactive machine learning education game where users train real algorithms, explore dataset geometry, and observe model behavior through playable visual experiments.
Next.jsReactTypeScriptTailwind CSSMachine LearningData Visualization
Project Overview
Gradient Descent Dungeon is an interactive machine learning education lab built inside the portfolio website. It teaches core ML concepts through playable levels, real algorithm behavior, decision regions, training metrics, and visual experiments.
Project Goal
- Teach machine learning concepts through interaction instead of static explanations.
- Show how algorithms behave on different dataset geometries such as linear data, XOR, moons, blobs, circles, and spirals.
- Demonstrate model assumptions, underfitting, overfitting, model capacity, ensemble voting, and clustering quality.
- Provide both guided lesson mode and free exploration through the playground.
Model Architecture
1.Next.js App Router provides the playable lab routes and level documentation pages.
2.Reusable React components render datasets, decision regions, metrics, controls, model reports, and playground views.
3.Custom TypeScript algorithm implementations run directly in the browser for Logistic Regression, Tiny MLP, Decision Tree, Random Forest, and K-Means.
4.Dataset generators create controlled synthetic datasets to demonstrate algorithm strengths and limitations.
5.A playground mode lets users freely combine algorithms, datasets, and parameters.