🚀 Get paid: Upload a question or refer a friend

1,230 questions reported asked in the last 12 months

Crack Your Next Interview

Practice the real questions OpenAI, Anthropic, Meta and Google actually ask — coding & LeetCode-style, system design, DL/LLM and classical ML — with instant AI review on every solution.

Real, reported questions from 345+ top labs & startups

3,285 questions · 2,713 coding & LeetCode-style · system design, DL/LLM & classical ML

Filter the whole bank to the last 6 or 12 months, or sort by most recently asked.

New

Sit a mock interview with an AI interviewer

It runs a real reported question the way an interviewer does — states the problem, lets you think, probes when you hand-wave, and won't give you the answer. At the end it files a debrief: a hire verdict, scores across seven dimensions, and what to practise next.

Included with Pro →

Straight from real loops

Real interview questions, every type

From system design to ML coding to flashcards — drawn from real, reported interviews. Here are a few of the most-viewed.

System Design for ML

Design recsys, serving stacks & data pipelines.

Coding & Leetcode-style

Implement attention, kNN, training loops from scratch.

ML Fundamentals & Algorithms

Bias-variance, gradients, classical ML.

Deep Learning & Architectures

Transformers, CNNs, normalization, optimization.

LLMs & Prompt Engineering

RAG, fine-tuning, evals & inference.

Forward Deployment Engineering

Ship ML with customers: integration, data & real-world debugging.

Flashcards & quick recall

Lock in fundamentals across every topic.

Everything in one place

Everything you need to prep — in one place

Scroll through what you get: real company questions, an in-browser coding pad with AI code review, flashcards, daily paper digests, an ML job board, and progress tracking.

OImplement multi-head attentionHard
ACode a transformer blockHard
MDesign a feature storeMed

Curated company questions

Hundreds of expert-reviewed ML questions, filterable by lab — Google, Meta, OpenAI, Anthropic and more.

Explore
def attention(Q, K, V):
scores = Q @ K.T / sqrt(d)
w = softmax(scores)
return w @ V
▶ pytest
✓ 3 passed in 0.21s

In-browser coding

Write and run ML code in the browser with LeanCode and the coding-practice editor — tests run instantly.

Explore
def two_sum(nums, t):
for i in range(n): for j in range(n): ...
AI Review

Correctness: allows i == j — likely the failing case.

Complexity: O(n²) — a hash map gets you to O(n).

New

AI code review

Get instant AI feedback on the code you write — correctness, complexity, code quality, and a targeted hint — without spoiling the full solution.

Explore
Flashcard

What problem does RoPE solve in transformers?

Reveal answer →

Flashcards

Quick recall decks across core ML, deep learning, and system design to lock in the fundamentals.

Explore
For you · arXivtoday

Mixture-of-Experts routing at scale

Long-context KV-cache compression

Popular

Personalized Paper Digest

AI-driven digests of new arXiv papers, tailored to your research interests — delivered daily.

Explore
O

ML Engineer, Inference

OpenAI · Remote

Apply
A

Research Engineer

Anthropic · Remote

Apply
D

Applied Scientist

Databricks · Remote

Apply

ML Job Search

Curated ML / AI roles from top labs and startups, matched to your profile and target companies.

Explore
This week
12-day streak · 48 solved

Track your progress

Attempt history, best scores, streaks, and per-company performance so you always know what to study next.

Explore

The honest pitch

Why not just LeetCode?

Generic algorithm grinding doesn't reflect how ML teams actually interview. Here's the difference.

Generic LeetCode
  • No idea when — or whether — a question was ever actually asked
  • Textbook puzzles with no link to any real interview
  • No idea which company asks what, or how often
  • Zero coverage of attention, RAG, training or serving
  • You get a green checkmark — never feedback on your code
Crack ML Interview✓ Verified
  • Dated: each question shows when it was last reported asked, and how often
  • Same algorithm and data-structure problems — but reported from real loops
  • Organized by company, so you prep the loop you're actually in
  • Plus ML system design, DL/LLM and classical ML in the same place
  • Instant AI review on correctness, complexity and code quality

AI code review

Solving it isn't enough. You have to show you can write great code.

Anyone can reach an answer. Interviews reward code that's clean, correct, and well-structured — and you only get there with real feedback. Every solution you write gets an instant AI review — correctness, complexity, and code quality — like a senior engineer reading over your shoulder, so your coding actually improves.

CorrectnessTime & space complexityCode qualityA targeted hint — no spoilers
Try AI review →