Attention Is All You Need (and So Are You)
A friendly, illustrated walk through scaled dot-product attention — why softmax, why the √dₖ scaling, and how multi-head attention lets a model look in many directions at once.
Math, machine learning & cute things.
Long-form notes with proper equations and illustrated covers — built on Material Design, with light & dark themes you can switch up top.
imagesmodecover
layersMachine Learning
imagesmodecover
layersMathematics
A Gentle Stroll Through Fourier Land
Sines, cosines and the surprisingly cozy idea that any signal is just a chord. We build the Fourier transform from scratch, with pictures.
imagesmodecover
layersMathematics
Eigenvalues, but Make It Cute
What does it mean for a matrix to "just stretch" a vector? An intuition-first tour of eigenvectors, spectra, and why they show up everywhere.
imagesmodecover
layersProbability
Bayesian Cats: Updating My Priors
A tiny cat decides whether it is dinner time. Along the way we meet priors, likelihoods and the posterior — Bayes' rule, gently.
imagesmodecover
layersOptimization
Gradient Descent at 3 A.M.
Learning rates, momentum and the small hours. A visual diary of rolling downhill toward a minimum without overshooting the valley.
imagesmodecover
layersMachine Learning
The Softmax Diaries
One function, infinite drama. Temperature, logits and why your model is suspiciously confident — a short, warm look at softmax.