# ClearDataLabs

> AI experiments and in-depth articles explaining neural networks, backpropagation, and machine learning from scratch — running live

ClearDataLabs is a non-commercial, open-source educational project by Kostiantyn Chumychkin. It builds AI systems from first principles in TypeScript/JavaScript with no ML frameworks, running entirely in the browser.

## Key Pages

- [Homepage](https://cleardatalabs.com/): Overview of articles and interactive demos
- [About](https://cleardatalabs.com/about/): Mission, philosophy, and author background
- [Articles](https://cleardatalabs.com/articles/): All technical articles
- [Projects](https://cleardatalabs.com/projects/): Interactive browser-based AI demos

## Article Series: Handwriting Recognition in the Browser (hwrjs)

A complete neural network for handwriting recognition built from scratch in TypeScript — no TensorFlow, no Python, no server.

- [Handwriting Recognition in the Browser, from Scratch](https://cleardatalabs.com/articles/hwrjs-handwriting-recognition-in-the-browser/): Series overview and live demo
- [Seeing in Cells](https://cleardatalabs.com/articles/seeing-in-cells/): How raw pen strokes become a 12x12 binary grid for neural network input
- [144 Numbers In, One Letter Out](https://cleardatalabs.com/articles/144-numbers-in-one-letter-out/): Building a 3-layer feedforward neural network from scratch in TypeScript
- [Backprop in the Browser](https://cleardatalabs.com/articles/backprop-in-the-browser/): Implementing gradient descent and backpropagation in JavaScript

## Article Series: Knowledge Extraction from Neural Networks

Methods to visualize and understand what a trained neural network has learned.

- [Extracting Knowledge Using Causal Index](https://cleardatalabs.com/articles/knowledge-extract-chapter-1/): Computing per-pixel causal index heat maps in pure JavaScript
- [Iterative Input Adaptation](https://cleardatalabs.com/articles/knowledge-extract-chapter-2/): Optimizing inputs to make the network "dream" — a lens on learned representations

## Live Demos

- [hwrjs Demo](https://cleardatalabs.github.io/hwrjs/): Draw a letter and watch the neural network classify it in real time
- [Knowledge Extraction Demo](https://cleardatalabs.github.io/knowledge-extract-ffnn-mnist/): Visualize what a neural network trained on MNIST has learned

## Source Code

- [GitHub: hwrjs](https://github.com/cleardatalabs/hwrjs)
- [GitHub: knowledge-extract-ffnn-mnist](https://github.com/cleardatalabs/knowledge-extract-ffnn-mnist)