No black-box libraries. Vectors, matrices, probability and gradient descent — implemented from scratch in Java, visualized live here.
A hand-rolled optimizer on the left, the real backend round-trip on the right — both served from a single Docker image.
x ← x − lr·f′(x) on f(x) = 0.18·x²
Static frontend → Spring Boot API → Postgres, all in one image. Each check does a real DB round-trip.
Pinging /api/db-test…
From the dot product to logistic regression — one concept at a time, each made to pass its own hand-written tests.
add · scale · dot · norm — vectors tip-to-tail.
multiply, transpose, inverse as linear maps.
eigenvectors and the decompositions behind them.
sample spaces, conditionals, Bayes' rule.
normal, binomial, sampling by hand.
estimators and maximum likelihood.
optimize a loss → linear regression.
classification + the first real library.