Netflix Software Engineer Interview Experience — Passed VO1 With a Billboard Rotation Coding Round and an ML Job Scheduler System Design

Netflix's VO (virtual onsite) has two rounds — VO1 and VO2.
VO1 is more general.
VO2 leans into stuff related to the specific team you're interviewing with.
I'll share VO1 here.

Coding: movie billboard rotation. Don't use a heap — use two treemaps. Two followups total: 1) multi-threading, 2) how you'd scale it in production and how you'd store/access the data.

SD: schedule ML jobs. The main points were:

1. What different kinds of ML jobs there are (training, batch inference, and scattered eval jobs — the focus is mainly on the first two)
2. What requirements apply when scheduling each type
3. What the scheduler's resource allocation policy looks like
4. How the scheduler monitors those compute resources
5. How each component handles crash handling
6. Touched briefly on multi-tenancy

Points 1-5 mattered most. The interviewer grilled each point, and he was really good at controlling the time — once you said something that hit the mark on a point, he'd pull the conversation back, which meant he'd collected the tech-depth signal he needed there, then he'd move to the next point. When prepping, you just need to be solid enough on points 1-5 that you can talk about each one for at least 5-10 minutes, then let him pull you back when he's ready.

The whole process went really smoothly. He interrupted me, but he was controlling the pace so that I could hit more points for him to collect signal on.

BQ: nothing much — just some classic collaboration, handling feedback, handling conflict.

Passed VO1.