Calculus · Linear algebra · Probability
For engineers who can train a model but can't yet defend it. Every derivation is shown in full, every figure responds to the parameter you change, and no step is left as an exercise for the reader.
No credit card required · 52 lessons readable immediately
What a lesson looks like
Each lesson opens as a chalkboard: the statement, the derivation, and a figure you can interrogate. Move a parameter and every dependent quantity updates with it.
Definition
limx→c f(x) = L
For every ε > 0 there is a δ > 0 such that whenever 0 < |x − c| < δ, it follows that |f(x) − L| < ε.
1. Choose ε. It is the tolerance on the output.
2. Find δ in terms of ε. It is the tolerance on the input.
3. Show the implication holds for every ε.
The syllabus
Start anywhere, but the sequence is deliberate: limits before derivatives, vectors before eigenvectors, probability before the objective functions built on it.
Interactive algorithm walkthroughs with runnable Python, step-by-step visualizations, and hidden test cases.
146 lessons
Production machine learning: MLflow, Kubeflow, Feast, and the pipelines that glue them together.
3 lessons
Batch and streaming data systems: Spark, Flink, and warehouse design patterns.
1 lesson
Modern frontend and full-stack development: Next.js, CSS Grid, and friends.
2 lessons
First principles through multivariable — limits, derivatives, integration, series, and calculus in several variables.
7 lessons
Vectors, matrices as linear maps, projection and least squares, eigen-structure, decompositions, and matrix calculus.
7 lessons
Foundations, random variables, limit theorems, estimation and inference, information theory, and stochastic processes.
9 lessons
Four phases: fundamentals, classical models from scratch, a GPT built end to end, then fine-tuning and retrieval.
4 lessons
Why it works
No "it can be shown that". When a derivation skips, the skipped line is the one you needed — so nothing skips.
Drag θ and watch the projection recompute. The picture responds to the parameter instead of illustrating one frozen case.
Every symbol is defined where it first appears. You should never have to search backwards to find out what a subscript meant.
Each topic ships beginner, intermediate, and advanced problems, so you can find the edge of what you know instead of guessing at it.
“I could train a model and not explain a single line of the loss function. Three weeks in, I derived backprop on a whiteboard in an interview.”
“The eigenvector lesson has a slider. I moved it, watched the basis rotate, and understood in a minute what a semester of lectures didn't land.”
“It never says "it can be shown that". Every step is on the board. That's the whole difference.”
Pricing
Every lesson is readable without paying. The paid tiers add graded problems and progress.
Read everything. No timer, no paywall mid-lesson.
For people working toward an interview or a course deadline.
Pay once. Includes courses that don't exist yet.
Comfortable algebra. The Pre-Calculus course rebuilds everything else from there — functions, trigonometry, logarithms — before Calculus I assumes any of it.
No. Figures recompute as you drag them, derivations expand step by step, and every equation is typeset rather than screenshotted, so it stays readable at any zoom.
That is exactly who it is for. The libraries hide the math until something breaks, a reviewer asks why, or an interviewer hands you a marker.
Any time, from your account page. Practice runs to the end of the period you paid for, then drops to Audit. Nothing you have read gets locked away.
Yes. Progress is tied to your account and never shown publicly or sold. Signing in exists to remember where you stopped, not to profile you.
The whole path is already on the board, in order, with nothing skipped in between.
Create your accountFree to read · No card required · Takes about a minute