Candidatei-me por indicação de um funcionário. O processo levou 4 semanas. Fui entrevistado pela Magic (San Francisco, CA) em mai. de 2025
Entrevista
Overall good clear experience, interview was not super hard. First there is a homework task, difficult but doable, then 2 rounds of interviews - first with engineers and then CTO
Take home around nuances of FP8. Was pretty interesting.
The next stage involved talking to the CEO, Eric.
He asked to derive the optimal algorithm for ring all reduce (popular in distributed training) .
It’s fair game to ask about domain specific specific things, but:
1) recruiter insisted nothing specialized would be asked, and there is nothing to brush up on ahead of time.
2) Eric also claimed the interview wouldn’t be specific to distributed training, which is odd given his choice of questions.
He seemed annoyed that I only had a non-optimal solution.
The main issue in this process is lack of professionalism. The recruiter mocked me several times, and hid important information.
Interestingly enough, the only thing they shipped thus far is marketing blog posts.
If you feel passionate about interviewing there, watch for the following:
(1) how close are they to actually shipping?
(2) is the recruiter showing lack of enthusiasm early on? Bail.
Brush up on distributed training essentials, and you’ll crush this.
Perguntas de entrevista [1]
Pergunta 1
Ring all reduce. Only optimal solution is accepted.
Candidatei-me online. O processo levou 2 dias. Fui entrevistado pela Magic em abr. de 2024
Entrevista
Got an email to provide a few more details about my application, before doing a call with the recruiter. Done in 30 minutes, very fair and pointed questions related to the specific of the job and my background.
Later I got a programming assignment which was fairly complex, but appropriate.
Perguntas de entrevista [1]
Pergunta 1
If I was world class level in this specific function.
