AI / ML Coding Challenges

Code the models,
from scratch.

Implement the architectures everyone talks about but few truly understand. Go from paper to working code — one challenge at a time.

...Builders
20+Problems
500+Submissions
attention.py — challenge #4
1import numpy as np
2 
3def attention(Q, K, V):
4  # scaled dot-product attention
5  d_k = Q.shape[-1]
6  scores = Q @ K.T / np.sqrt(d_k)
7  w = softmax(scores, axis=-1)
8  return w @ V
9 
10def softmax(x, axis=-1):
11  e = np.exp(x - x.max(axis, keepdims=True))
12  return e / e.sum(axis, keepdims=True)
Transformers
Diffusion
RL
LSTMs
CNNs
VAEs
Graphs
Attention is all you need
No black boxes
Paper to code
Implement from scratch
Build real intuition
2400+ builders
Read the paper
Write the math
Pass the tests
Level up
Attention is all you need
No black boxes
Paper to code
Implement from scratch
Build real intuition
2400+ builders
Read the paper
Write the math
Pass the tests
Level up
The Journey

From paper
to production.

Most people use models without understanding how they work. buildml changes that — by forcing you to write every line yourself, you build genuine intuition no tutorial can give you.

01
Read the Paper

Every challenge begins with the original research. No hand-holding — read the source, understand the theory.

Foundation
02
Implement Cold

Write the code from scratch in our editor. No copying, no shortcuts. Just you, numpy, and the math.

Practice
03
Run the Tests

Automated checks validate your implementation against reference outputs. Debug until it passes.

Verify
04
Level Up

Climb the leaderboard, unlock harder challenges, build a portfolio of real ML expertise.

Grow
Zero to Infinity

You start with basic challenges, gradually master the fundamentals, build real-world projects, and eventually finish by producing proper machine learning research papers.

Test-driven

Each challenge ships with a full test suite. You know you're right when all tests pass.

Paper-first

Every challenge is grounded in a real research paper. Understand the theory before you type a line.

Feedback

Help us improve.

Your feedback helps us improve buildml. Share your thoughts, report bugs, or suggest new features.

Stop using black
boxes.

The best ML engineers don't just fine-tune APIs. They understand the math, they've written the code, and they know exactly what's happening inside the model. That's what buildml builds.

Start for free →