Machine learning and AI

Tiny CNN walkthrough
Follow a 6×6 image through convolution, ReLU, pooling, flatten and softmax, edit any weight, and train the network yourself.
Tiny LLM walkthrough
Follow a sentence through tokens, embeddings, positions and self-attention to a prediction of the next word, and edit any weight.
Log autoencoder walkthrough
Find anomalies in a real SSH auth.log: follow one line through an embedding and Conv1D layers, compare four tokenizers, see what the model rebuilds and what it can't, and set the threshold yourself.
Image recognition walkthrough
Husky or malamute? Follow real photos through filters and feature maps, compare a CNN learned from scratch with transfer learning, see what each model looks at, and try your own photo.

Other useful tools

Made by others, and worth exploring next.

Transformer Explainer

Runs a real GPT-2 model in your browser and shows how your text flows through it: embeddings, attention heads, the MLP and next-token probabilities, with a temperature control. The same steps as the Tiny LLM walkthrough, at full size.
poloclub.github.io · Polo Club of Data Science, Georgia Tech