Mistral AI Software Engineer Interview Experience — ML Fundamentals for a Research Role

The author reports an unsuccessful interview for a research-engineering role at Mistral. After an introduction to the team, the applicant spent several minutes discussing relevant background and projects. The interview then moved through direct questions about model parallelism, mixtures of experts, normalization, diffusion, optimization, preference-based training, and keeping GPUs supplied with data.

The discussion emphasized distinctions between related techniques rather than an extended coding exercise. When the applicant asked about company direction and competition, the interviewer described a focus on enterprise customers and business-facing AI services. The author states that they could not answer the diffusion and Adam questions and were rejected. The report does not describe later rounds or provide a separate employer explanation for the decision.