KnowML
Machine learning and modern AI, from the maths up
Start anywhere and follow what interests you. Linear algebra through to agents, and the systems work underneath.
How everything here is written
Six questions, asked of every model, in the same order
A decision tree and a diffusion transformer get the same interrogation. Learn the shape once and every new architecture becomes six familiar questions instead of one unfamiliar diagram.
Representation
→
Objective
→
Architecture
→
Training
→
Inference
→
Deployment
Four ways in
Pick the one that matches where you are
I don't know where to start
Roadmaps
Four ordered paths with honest time estimates, a checkpoint you must pass before moving on, and — rarest of all — what to skip.
Choose a path →
I want to see it actually run
Labs
Step through attention, backprop, BPE, quantization and sampling one operation at a time, on a worked example small enough to check by hand.
Open a lab →
I need a model for a job
Model Atlas
Language, speech and vision models side by side — open-weight first, with licences, benchmark numbers and what each benchmark measures.
Compare models →
I have an interview on Thursday
Drill & practise
310 interview questions on a spaced-repetition schedule, plus a map of where to go write the code — Deep-ML, Kaggle, MLContests and the rest.
Start drilling →
What’s actually in here
The whole field, not a greatest-hits tour
Nothing here starts at the transformer, and nothing stops at it either. The parts that usually get skipped — CUDA kernels, serving economics, evaluation, what the industry results really showed — get the same treatment as attention.
While you learn
The site is a workspace, not a document
Search everything
⌘K for topics, ⌘F for this page. Every section, subtopic and term is indexed client-side.
Highlight & annotate
Select any sentence to highlight or note it. Autosaves to this browser, exports to Markdown.
Diagrams that move
Pausable data-flow diagrams: attention weights, residual streams, denoising steps.
Runnable examples
Short snippets you can copy, change a number in, and see what moves.
37
Sections, all written
310
Interview questions
249
Techniques mapped
5
Interactive labs
Start anywhere
Every section grouped by theme, the 1950s-to-2026 timeline, the dependency order, and where to go once you have read it — on one page.
Browse everything → 🌳 Technique Map