I was contacted by a recruiter about a Senior Solutions Architect position. They were looking for a search expert, and the role sounded like a perfect fit for me. I completed an hour-long assessment, which included a personality test, a lengthy math test, and a vocabulary test. Afterward, I met with the hiring manager and someone in a similar position for a technical interview. They had allocated an hour for this interview, but the interviewer ended it after only 20 minutes.

I was very surprised to be turned down after the technical interview. The stated reason was a lack of experience with vector search and RAG, despite none of the interviewers asking a single question about these topics. In fact, no one asked me any technical questions during the interview process, aside from requesting I describe my experience.

I contacted the recruiter and explained that my interview was cut short and that no questions were asked about the subjects for which they claimed I lacked experience. To their credit, they then scheduled another interview with a more senior engineer in the department. However, we only had 30 minutes to talk, making this interview very rushed and exceeding the allotted time.

This interviewer focused solely on complex AI questions. This made discussion difficult in a rushed setting, as companies use varying terminology for this new technology. I had watched some YouTube videos beforehand to familiarize myself with their approach, but it didn't significantly help as different MongoDB employees used slightly different terms.

This time, they acknowledged my knowledge of vector search and RAG. However, I was ultimately rejected for insufficient experience in creating LLMs and text embeddings. In my previous role, I was the primary troubleshooter for text embeddings created by other engineers. I have also developed other types of Gen AI models and used several LLMs. The job description only required three years of experience with searching databases and software development, with no mention of AI experience. It felt like the goalposts were constantly shifting, and they were unsure of their requirements.

I have almost 20 years of experience in tech and worked on vector search for a few years, starting five years ago. It was baffling that this was not considered sufficient. The requirement for experience creating LLMs, which is not part of the role and not something I would need to do, was particularly unexpected. The job description highlighted learning about vector search as a perk, with no mention of RAG, building LLMs, or embeddings. By the time of my interview, the position seemed entirely different.

During the first technical interview, all questions were general, such as, "What is your experience with scripting/databases/AI/Networking/Security/etc.?" There were no technical or follow-up questions to gauge the depth of my knowledge. This interview process would be very easy to navigate if one were misrepresenting their experience.

The second technical interview was extremely complex. I faced numerous questions that forced me to consider my past NDAs, such as specific details about how my previous company utilized vector search, including embedding processing and exact specifications like the number of dimensions used. The interviewer seemed to be using "dimensions" in a non-standard way, not referring to the number of vectors in a dense vector array, though I was unsure of their intended meaning. He also asked many questions unrelated to the job, like my use of Stable Diffusion (an image AI). At times, it felt like he was soliciting advice for his personal AI projects. I initially considered offering solutions based on his described home setup, but then questioned its relevance to the position. None of the questions asked pertained to the job description.

It was amusing that the first interviewer mentioned having ample free time and limited daily tasks, while the second interviewer described being constantly busy and on the phone, emphasizing the need for more staff, despite both holding seemingly identical positions within the company. Instead of requiring candidates to undergo two management interviews (the next would have been with the VP, followed by a presentation), they should conduct two in-depth technical interviews. This would allow sufficient time to cover essential topics like search and AI and enable candidates to meet more team members. This discrepancy in their accounts was a significant red flag for me and diminished my interest in the position.

Furthermore, the advertised salary range is considerably low and does not align with the level of prior AI knowledge they expect.

How does Stable Diffusion use CLIP?

How would you solve a problem with a chatbot model where it thinks it can write a book but is making no actual progress on writing that book and instead just tells users that it's working on it?

How is RAG used and what problems does it solve with a chatbot?