All tracksStart here 30 60 200
Generative AI Challenges
Learn the idea, implement the important part, then prove you can use it.
Your challenge path
0/3
Challenges complete · progress is saved in this browser
The curriculum
Build the skill one module at a time.
The first two levels are untimed. Later challenges add pressure only after the workflow feels familiar.
Module 1
Generative AI Foundations
tokens · probabilities · sampling
Choose your difficulty
All levels
Browse the full module path
Guided
Whitespace tokenizer
Implement `tokenize(text)` returning lowercased word tokens with punctuation stripped.
No timerOpen
Supported
Softmax sampling
Implement numerically-stable `softmax(logits)` returning a probability distribution.
No timerOpen
Interview
Scaled dot-product attention
Implement single-head self-attention: softmax(QKᵀ/√d)V. Match the reference output.
Timed when startedOpen
Understand the idea first. Then make the code yours.