I mass-applied online and roughly two weeks later got an email from a recruiter, scheduling a recruiter phone interview.
Recruiter phone interview
This round mostly asked about my related background (my experience doing LLM post-training), then three behavioral questions (an end-to-end work experience, a time I failed, a time I disagreed with someone). The recruiter said the team works on Physical AI and could offer a Senior Applied Scientist position. They also said I didn't need a Physical AI background — a basic post-training background was enough.
Technical phone interview
About a week later, the recruiter scheduled the technical phone interview for another week out. This round was an hour, structured as follows:
First a 5-minute self-introduction.
Then about 15 minutes talking about the background of projects I'd worked on.
Then a coding problem. The task was to build a complete data pipeline, including pad_vector_collate, create_data_iterator, and run_data_pipeline, plus support for batch dropping. The test cases included some corner cases. The coding portion took 20 to 25 minutes.
After the coding, there were 10 minutes of rapid-fire LLM questions, mostly concept-level — things like: explain the training objective of SFT, the difference between DPO and PPO, the difference between GRPO and PPO, why many cases still need SFT and DPO, what DPO is actually optimizing, why DPO doesn't need a KL term, the pros and cons of GRPO, why GRPO's reward signal counts as a non-dense signal, when to use LoRA-type methods for fine-tuning, what the problems with LoRA are, and so on.
Reverse-interview and wrap-up
In the last five minutes I got to ask questions. I asked what the team actually works on, but I felt they weren't that willing to go into much detail — they just said it's modeling related to Physical AI, also post-training, using open-source models.
Overall the interview experience felt pretty average.
Discussion
Loading comments…