J'ai postulé en ligne. Le processus a pris 3 semaines. J'ai passé un entretien chez Canva (Melbourne) en juin 2024
Entretien
The interview was a hr screening followed by a coding interview with a senior ML engineer.
The hr screening was smooth, he went through my experiences and asked few questions about my familairity with certain technologies.
The coding challenge was new in its nature though. Unlike the standard leet-code questions where you are given a problem and should write a code for it (the style I was more ready for), for canva, you had to work on an existing code and complete different parts; which was good and bad.
The experience was new, but the fact that I was thinking from scratch was not meant for this type of coding interview. Instead, you should have spent some time reading the question and the existing code and then approach the problem.
The interviewer was not very helpful with asking me to pay more attention to existing code instead of trying to write new ones.
Questions d'entretien [1]
Question 1
Questions were around confustion matrix and clustering alrgoithm.
3 interviews.
1. Hr call to initial screen
2. "Vibe coding" interview that uses AI tool
3. Interview loop for 3 hours
Including behaviour, ML design and coding round without AI.
HR screen where they asked about experience particularly if you have worked with TB or PB data processing pipelines.
Then an AI programming interview where you need to remove gender bias from some recommendations.
After this it's 4 more 1 hour interviews. ML System Design, ML System review, Pair programming without AI and a behavioural interview.
I also met an Engineering Manager to discuss the role which he described as primarily a software engineering role and that they don't do any model training.
Each interview only had one interview which in one instance the interviewer had pretty poor communication skills that caused confusion and wasted quite a bit of time. It also means that the interview outcome was subject to bias as there were no other interviewers to align with.
Most disappointing though, there was no meaningful feedback provided that I could take away from the experience and build on.
AI assisted interview for first round and create a model without AI second round, system design and then culture fit as subsequent rounds. First round preferable is code helper like GitHub copilot or Claude code
