The recruitment process was hindered by scheduling issues. Every stage was cancelled and rescheduled multiple times, and in one instance, no interviewer showed up for the scheduled slot.

The technical stages consisted of:

• A code review
• An "LLM quiz"
• A live coding session

The LLM quiz covered standard topics: transformer architectures, fine-tuning, and RAG, but the format was a rigid Q&A rather than a technical discussion. It felt as though the interviewer was looking for specific keywords from a prepared sheet rather than exploring depth of understanding.

The live coding stage involved using an external API, but was difficult to navigate due to interviewer wifi problems. They joined 10 minutes late with a poor internet connection, which led to communication gaps and frustration when I asked for clarification on the task. The goals of the exercise were not clearly defined, and overall it was hard to understand what was expected.

I received a rejection two weeks later via an automated email. The feedback suggested I "work on my deployment skills," which was confusing as deployment was not a component of any of the interview stages. It feels like a shame; I think Mistral may lose many good candidates with a process like this.

How do you evaluate RAG?