Prep built
for frontier AI interviews

Practice with real questions candidates were asked, with access to the largest AI interview database. Write, speak or code, then get line-by-line feedback.

Start preparing →See what you get
1,716
Real interview questions
78
AI companies
198
Practice challenges
3/3
tests pass
Seen at Lyft
ML / AI Engineer · ML Fundamentals
Implement K-means centroid updates
Medium20-40 min
solution.pyPython 3.13 Autosave on
def kmeans_centroids(points, centroids, iterations):    """Return centroids after the requested rounds."""    cur = [list(c) for c in centroids]    for _ in range(iterations):        buckets = [[] for _ in cur]        for p in points:            i = min(range(len(cur)), key=lambda j: d2(p, cur[j]))            buckets[i].append(p)        cur = [mean(b) if b else c for b, c in zip(buckets, cur)]    return cur
Test casesran in 0.42s
Two clear clustersPASS
Tie goes to lower indexPASS
Empty cluster held unchangedPASS
Run checks these. Submit also checks hidden cases.
Preparing candidates for roles at
+72 more

Built for how frontier labs evaluate candidates

Questions from real interviews, scored against a role-specific rubric.

345ScreenPhoneTechnicalOnsiteFit

Know what's coming

The stages, formats, and questions for the exact company and role you're going for. Know what to prepare and what to ignore.

01:24:17

Rehearse how they'll test you

Write the design doc. Code against real tests. Answer behavioural questions out loud. Same formats, same pressure.

Problem framing7/8Technical rigor6/8Risk awareness8/8Clarity5/8

Find out before they do

Every answer scored against a role-specific rubric, with the reason behind each mark. See what's strong, what's thin, and what to fix next.

How it works

Make the interview the easy part

Because by the time you sit it, nothing is new.

See what you're walking into

Fresh reports reveal what each stage tests

  • ✓Role-specific interview maps across frontier AI labs
  • ✓Updated from real candidates' recent interviews
  • ✓Detailed guides to each interview round
  • ✓Common pitfalls and preparation priorities
  • ✓Relevant practice questions and challenges
Start preparing →Browse tracks
Research Engineer · Frontier Lab
Active
Stages
6
Duration
8wk
Progress
62%
✓
Recruiter screen
Day 1 · 30min
✓
Coding screen
Day 9 · 60min
3
System design screen
Day 20 · 60min
4
Onsite coding
Day 45 · 60min
5
Onsite system design
Day 45 · 60min
6
Behavioural interview
Day 52 · 45min

Practice on real tasks

Practice beyond question lists and quizzes

  • ✓198 challenges calibrated by track and seniority
  • ✓Technical memos, advanced coding, system design and behavioural questions
  • ✓New reports added every week
  • ✓Answer by voice on most challenges
Try a free challenge →
Prompt
Notes
Rubric
01:24:17
CHALLENGE 047 · RESEARCH ENGINEER
Design a safety evaluation protocol for a general-purpose language model prior to public release. Identify the three failure modes you would most want to catch.
1. Overview
The protocol targets three failure modes ordered by expected frequency of real-world harm.

2. Reward hacking under deployment drift
We probe the model against a held-out set of
412 words · 6 paragraphs · 78 wpmSubmit for grading

Your performance in detail

Scoring and feedback tailored to each challenge

  • ✓Feedback returned in minutes, not days
  • ✓Strengths, weaknesses and missed points
  • ✓Progress tracked across submissions and revisions
  • ✓Actionable feedback after every submission
  • ✓A clear verdict on your performance
Try a challenge →
Rubric Report
Alignment & Safety · Frontier AI lab
Evaluation protocol for a frontier model
71/100
In top 42%
Problem framing
3/5
Framing is present but the risk model is stated rather than motivated.
Technical rigor
4/5
Method choice is defensible; baselines cited appropriately.
ML-specific reasoning
3/5
Underweights how RLHF reward signal degrades outside training distribution.
Evidence & specificity
3/5
No mention of red-team budget tradeoff. Senior reviewer would flag.
Clarity & communication
3/5
Key recommendation is buried mid-document. Move up for reviewers.
2Strengths
4Weaknesses
3Missed

What our users say

Being able to retry the same challenge made a real difference. I could fix what I'd missed and see my scores improve. Eventually I started catching those mistakes myself.
Alex
I had two months to prepare and needed to make them count. The company guides helped me work out what to study for the labs I actually wanted to join.
David
I liked having so many relevant questions in one place. The recent interview reports were especially useful for seeing what companies are asking now and deciding what to practise.
Min

Feedback collected from pilot users

Prepare for the role you want

Choose your role to find relevant questions and challenges.

Software Engineering

Coding, system design and infrastructure

Machine Learning

Models, inference and evaluation

Research Engineering

Experiments, model training and evaluation

Product Management

Product strategy, decisions and metrics

Forward Deployed Engineering

Customer problems, solution design and pilots