Experiments
Small, self-contained demos I build to understand how models and systems really work. The computation is real and runs in your browser. Click any one to play with it.
A real transformer with hand-written backprop, trained to copy from context. Watch the attention arc jump back to the matching token. Gradients verified against finite differences.
Train a small neural network from scratch to classify 2-D data and watch the decision boundary form. Backprop and Adam by hand; pick a dataset and tune the learning rate.
A generative adversarial network training in real time: two hand-written neural nets (backprop and Adam from scratch) fight until an amber cloud snaps onto a ring.
Scaled dot-product attention you can grab and move. Drag the query vector and watch the softmax weights and the attended output update live.
A real query engine in the browser: tokenizer, recursive-descent parser, and executor. Run SELECT / WHERE / ORDER BY and inspect the tokens and syntax tree.
Drop a ball on a 2-D loss surface and tune the learning rate to watch it converge, oscillate, or diverge.