I come from finance and wanted to move into AI and LLM work. For a few months I took on projects with early-stage startups, building RAG systems and AI-native products on the side.
I want to stress one thing, because I didn't really believe it myself before: you can switch fields even in the middle of a career, with no big tech background. I had no referral either and just applied cold. What got me noticed was building actual systems and being able to explain them properly. Side projects and hands-on work count for a lot if you're trying to make the same jump. Funnily enough, the thing I kept struggling with while building was how to evaluate LLM outputs, and that turned out to line up closely with what this role was about.
Interview Process
- Initial Round
- Data science style coding (EDA, then modeling, then evaluation)
- A chat with the hiring manager
- Loop Round
- ML System Design
- Coding
- AI Exercise (LLM evaluation plus a bit of implementation)
- One more System Design Round
Preparation
I had roughly two and a half weeks before the loop and put in around 5 to 6 hours daily, splitting the time between system design, LLM evaluation concepts and coding. What helped most was making practice feel like the real thing: I set a timer and said my answers out loud rather than just running them in my head. I also liked dictation mode a lot more than voice mode for this, since voice mode cuts in on you and dictation lets you ramble and think at your own pace. For coding I did LeetCode easy and medium problems, mainly to stay relaxed writing code under pressure, even though the real coding round turned out to be more practical. My system design prep at first was mostly on my own, talking through LLM-based systems and putting together a rough framework. On the AI side I concentrated on evaluation, detecting hallucinations, and the ways these systems fail.
Interview Experience
System design was the first round of the loop and it was the hardest for me. The question was about evaluating and improving LLM systems. I'd come across pieces of it before but never prepared it in any organized way. I jumped straight into answering instead of pausing to pin down the problem. Partway through, the interviewer told me to draw the system since my verbal explanation wasn't getting through, and doing both at once, explaining and sketching live, was harder than I thought. I knew right away that it hadn't gone well.
The coding round came next and felt much better. It was more practical than the usual LeetCode style and was about how to structure code and think about real-world issues like versioning and consistency. The AI exercise was the following day, and that's where the prep paid off. We dug into hallucination detection, evaluation approaches and how to measure LLM performance, and I also had to implement a small component.
Around 3 business days afterwards the recruiter got in touch and asked whether I'd do another system design round. I wasn't expecting that after how the first one went, but it showed me that a single bad round doesn't automatically end things. This time I changed how I prepared. I spent my time learning the system design fundamentals through an online course, which helped me think in patterns and break problems down in a more orderly way. I also practiced drawing systems while talking, and did a mock interview too.
The last round was about designing a memory layer for an LLM system, something that could pick up and reuse information from earlier interactions. About a week after that, I got the offer!!
Key Takeaways
My main lesson is that LLM interviews are less about landing on the correct answer and more about how you deal with vague, open-ended problems.
A few things that stuck with me:
- A bad showing in one round doesn't mean the offer is gone
- Practice system design by talking through it out loud, a lot, because it feels nothing like thinking it through silently
- Do at least one mock interview, since it's the nearest thing to the real setting
- Ask your recruiter for as much detail about preparation as they'll give you, because it helps you focus
Discussion
Loading comments…