went through the full Mistral AI software engineer interview process earlier this year. applied for a senior backend role, Paris-based but interviewed remotely. sharing everything because I couldn't find much detail when I was prepping.
timeline was pretty tight. recruiter reached out within a week of applying, first call happened about 4 days after that. from first contact to offer was roughly 5.5 weeks.
the stages: recruiter screen (30 min) - standard stuff, motivations, availability, compensation range technical phone screen (60 min) - one coding problem, medium difficulty, plus a short chat about past projects take-home (3-5 days) - not huge, a focused implementation task, they gave clear time estimates, I spent about 4 hours full onsite loop, 4 rounds same day: system design (60 min) deep dive on take-home with the engineer who reviewed it second coding round, more open-ended behavioral / values (45 min)
total investment: probably 8-10 hours of actual time once you count prep.
the coding rounds were harder than, say, a typical fintech but not LeetCode-hard in the grinding sense. more about actual engineering judgment. they wanted to see how you think about trade-offs, not just whether you can implement Dijkstra in 20 minutes.
the take-home was the most interesting part. you submit it and then someone walks through it with you live, asking why you made specific choices. be ready to defend your abstractions. I got pushed pretty hard on a caching decision I made.
overall vibe: small-company energy, people seemed genuinely curious about the tech rather than going through motions. not a lot of HR polish but the process was respectful of time. they do move fast if they're interested.
one thing I didn't expect: the system design round leaned heavily on infra and scalability for AI/ML serving, not just generic distributed systems. knowing about inference pipelines, model serving latency, and GPU memory constraints helped a lot.