OpenAI Machine Learning Engineer Interview Experience — Onsite Dragged to November, Then Rejected

I applied in June, and they didn't reach out to schedule an interview until September. The phone screen went well, and they quickly set up the onsite (VO), but from there the scheduling turned into a mess — they kept changing the time on me, over and over, and it dragged all the way to November before I finished every round. Two weeks after that, I got the rejection.

I won't go into every detail, but here's roughly what came up in each round:

Phone screen

• ML Coding: mainly tests how familiar you are with the Transformer architecture — you need to be able to rattle off the basic concepts smoothly.
• ML Design: a recommendation & search related design question.

Onsite (VO)

• BQ: standard behavioral questions. My personal read is that they care about two things most: why do you want to work at this company, and how aligned your personal mission is with the company's mission.
• Coding: I think it was two main parts plus a bonus. Because I ran out of time, I only managed to finish the first two parts, and the interviewer just didn't let me attempt the bonus. The problem itself wasn't that hard, but it took a while to figure out the pattern.
• ML Coding: you really need to know numpy cold — all sorts of operations need to come out of your fingers without thinking.
• Tech Deep Dive: this was mostly the interviewer digging into things I'd written in my application materials. It felt like whatever you'd said, they'd grab onto it and keep pushing further.

Overall: honestly, the whole process was a pretty bad experience — it felt kind of like a disillusioning process.

The interviewer from the phone screen left me with a pretty good impression of the company. They seemed like a decent person and the conversation flowed well.

But once I got to the onsite, the experience dropped off a cliff: the scheduling was chaotic, they'd change the time on short notice, and some of the interviewers were pretty phoned-in too. On top of that, they had people with more of an SDE background interviewing the RS tech deep dive, and it just felt off talking to them. The process dragged on way too long and the communication was bad — whatever rose-tinted glasses I had for this company disappeared instantly.