ML / Data · expert
PyTorch
Tensor mechanics, autograd, training loop discipline, mixed precision, debugging gradient issues.
What to expect
- 1
Pre-flight (~ 1 min)
Quick selfie + mic/camera/network check. Same setup as every other Winkle interview.
- 2
Conversation (~ 25–35 min)
Winkle asks questions one at a time, in your language. She probes weak answers with one follow-up, then moves on.
- 3
Coding block (~ 10 min)
A small live coding task in a built-in editor. You explain your reasoning as you go.
- 4
Verdict in ≤ 30 seconds
Winkle writes a scoring artifact: 7 per-question scores, strengths, concerns, and a verdict. You see it instantly.
Sample question
“Your gradients explode on epoch 3. Walk me through your debugging tree.”
Topics covered
Before you start
- Find a quiet space. Wired audio if you have it.
- Good lighting on your face. Avoid backlight.
- You can take the interview in any of 8 languages.
- Block 40 minutes. There's no pause button.
Why this matters
Verified skills surface you to recruiters before unverified candidates. Each badge is valid for 12 months — retake any time to refresh.