Interviews now let
you use AI.
The real skill is driving_it_well
VibeGrill gives you realistic Python tasks, an AI pair to work alongside, and a score on how well you collaborated with it — not just whether the tests passed.
# fix the off-by-one at the window edge from collections import deque def rate_limit(hits, window): seen = deque() for t in hits: while seen and t - seen[0] >= window: seen.popleft() seen.append(t) yield len(seen)
What does >= include that >wouldn't — the hit sitting exactly window ago?
▸ level 1 · nudge — no code yet
// how it works
Three steps, one browser tab
Pick a realistic task
Fix a bug or build a feature in a small Python codebase — or take on a real Meta / Google interview problem.
Solve it with an AI pair
Chat for advice, or hand the wheel to an agent that edits your files. Everything runs right in your browser.
Get scored on how you drove
Hidden tests check correctness. An AI judge reviews the whole session and rates how well you drove the AI.
// what gets scored
We don't grade the code.
We grade how you drove the AI.
Anyone can paste a prompt. The skill worth practicing is everything around it. After each session an AI judge reads the full transcript and your diff, then scores five dimensions — the ones that actually separate a strong operator from a copy-paster.
- 01
Problem decomposition
Did you break a fuzzy task into steps the AI could actually act on?
- 02
Prompt quality
Did you ask for the right thing, with the right context, precisely?
- 03
Verification
Did you run it, test it, and probe the output — instead of trusting it?
- 04
Catching AI errors
Did you spot the bug the model introduced with total confidence?
- 05
Independence
Did you steer the work, or did the AI quietly end up driving you?