I had 3 interviews with different engineers/scientists, totaling 3 hours.

I applied for 2 different roles: Research Scientist and GPU Engineer. Due to very poor communication with the recruitment team, I am unsure precisely which role I passed interviews for.

Communication with the recruiter was limited to email.

The first interview was with a GPU/performance engineer. It was a pleasant discussion about my background, followed by questions on software performance.

There was no news for over a week. I had to inquire about my status to learn I was progressing to a second round, which was apparently for a different role.

The second interview was with an AI/research scientist. This also involved a discussion, with questions such as how to test AI models for regressions before they go into production.

The third interview was with an AI scientist and included a coding round with two parts:

• The first part involved coding a function in PyTorch. Given the coordinates of points and the coordinates of centers, the function was to compute the distance to the closest center for each point (using broadcasting and torch.argmin). All inputs and outputs were torch tensors. This is a component of the k-means algorithm, but no specific details about the algorithm were requested.
• The second part involved parsing a string composed of numbers and '+' or '*' signs to compute the result, akin to a calculator without parentheses. I found it to be a good coding round, given the allocated time.

The interviews with the engineers and researchers were good and quite challenging. However, the communication with the recruitment team was very poor.

Two weeks after the last interview, with no updates, I contacted the recruiter. They informed me they did not wish to proceed, offering minimal feedback that I felt did not align with my experience. They did not reply to my subsequent emails.

I have never been treated so poorly by a recruiter before. I hope Mistral significantly improves its recruitment process, as this is unacceptable for a leading tech company recruiting highly qualified talent. If this continues, it will eventually cause problems.

Coding a function with PyTorch that, given the coordinates of points and the coordinates of centers, computes the distance to the closest center for each point. This utilizes broadcasting and torch.argmin. All inputs and outputs are PyTorch tensors. This is a part of the k-means algorithm, but no further details were requested.