
Machine Learning Engineer Interview Experience
Interview process
technicals were passable with some prep. for onsite, it helps to be familiar with more exotic forms of pipeline parallelism
- Technical interview
- Final round
Interview rounds
2 questions across 2 roundsTechnical interview
1 questionSpecific questions asked
implement variants of spec decode
Final round
1 questionOverview
Implement it with PyTorch in a codebase
Specific questions asked
implement variants of pipeline parallelism
Interview tips
standard transformer decode logic in pytorch for technical. for onsite, read up on pipeline parallelism implementations, in particular ones that are more sophisticated, other than the standard gpipe.
Company culture
onsite was more like a realistic project, was quite interesting
Get full access with a membership, or share your experience to try it free.
More Cursor interview experiences

“I had to lead an enablement session for a thousand Adobe engineers, so I built a fake Adobe codebase, demoed Cursor live, and got pushback like, "usage is up, but PRs are going down, what do you do?"”

“They basically gave me access to part of their codebase and were like, "Figure it out, see anything you want to build, and just build it." For eight hours I worked out of a Slack group, then presented the feature at the end of the day.”