Every problem here is one you implement yourself. No library call standing in for the thing you are trying to learn. Real Python, real compute, free.
free · no account to try
fill the blank, then Run
Numerically stable softmax · runs on deep-ml compute
the max subtraction is what stops exp overflowing · problem 23
700+ universities represented
engineers at these companies practice here
OpenAI, Anthropic, Google, NVIDIA, Meta, Apple, Microsoft, Amazon, Tesla, Netflix, Stripe, Mistral AI, ByteDance, Character.AI, Together.ai, Airbnb, LinkedIn, Coinbase, Spotify, JPMorgan, Morgan Stanley, Tencent, Qualcomm, Intel, Samsung, Atlassian, MongoDB, Visa, Capital One, Revolut, BMW, Airbus, Bosch, Accenture, Deloitte, Flipkart
students at 700+ universities practice here
Browse and solve machine learning coding challenges. Filter by difficulty and category, and track your progress across the catalogue.
open problemsloading the catalogue
click a difficulty
An ordered route through the problems, labs and projects, built in the order they make sense. Each path is bound as a book, and beginner through advanced is inside it rather than split across three separate courses.
open learning pathtaking the books down
take one down
Pick a company, follow a paced path, build the resume projects they screen for, run timed mock interviews, and track where you are.
open interview preppaced path · resume projects · timed mocks
pick a company
Meet Zero. Ask him anything and he answers from the real catalogue: he recommends actual problems, builds a playlist you save with one click, and remembers what you are working on. On a problem page he reads the question and your editor code, so he is tutoring you on the thing in front of you.
open ai assistantwaking the assistant
tap a starter to ask him something else
Implement an algorithm and test it on an actual dataset, with constraints, evaluation metrics and time limits. No CUDA install, no environment to break.
open labsloading labs
open a lab
An atomised series of problems that compose into a complete build. Solve every step and run the model you built.
open projectsloading projects
build one
Each problem opens with a worked example, then three practice problems to build mastery. Inspired by Math Academy.
From the four fundamental subspaces to PCA.
open mathloading math
work through it
Collections of problems organised by topic. Track your progress through each one and earn a badge when you finish it.
open collectionshover a card
Build a playlist of problems and labs into a study plan, then share it. Copy anyone else's into your own library.
open playlistsloading playlists
drag to reorder
Timed coding competitions against other practitioners. New problem sets, live ranking while the clock runs.
open contestsa sample · see the real ones
The top practitioners on Deep-ML. Climb it by solving problems and finishing labs.
open leaderboardloading the board
drag your heat
The crowd-trained tiny LLM. Fork one slot of the canonical model, improve it, and if you beat the hidden test metric your code merges automatically.
open researchloading the challenge
every dot is a submission
Every figure here is the real computation, running in the page.
Ax drawn against x. Two directions survive.
practice thisFree to start.
run your hand through it