← Back to experiences
[ OK ] Loaded —
[ INFO ]
$ cd
$ ls -lt
01
02
03
04
05
$ ls -lt
01
02
03
04
05
user@intervues:~/experiences/…$
Level: Mid-Level
Round: Onsite · Type: Multiple Types · Difficulty: 7/10 · Duration: 240 min · Interviewer: Unfriendly
Topics: Presentation, Live Coding, LLM, MySQL, DynamoDB, NLP, System Design, Behavioral
Location: San Francisco Bay Area
Interview date: 2025-03-15
Got offer: False
My onsite interview consisted of four rounds.
Round 1: Introduction & Craft Demonstration I was asked to prepare a 15-minute presentation, including a 5-minute introduction and a 10-minute 'proud project'. I also needed to set up a repo on GitHub. During the interview, after my 15-minute presentation, I had 75 minutes of live coding while being observed. The interviewers did not seem prepared and hadn't reviewed my repo's code beforehand. When I asked what they wanted me to modify, they said to do whatever I wanted. I chose to work on an unfinished function in my repo, which involved generating a description of a player based on player information using an LLM. The process involved feeding a prompt to the LLM to generate a SQL query, using the query to find player information in a database, and then feeding this information back to the LLM to generate a natural language description. Since it was my first time writing Python backend code, I had to look up various APIs (and I wasn't allowed to use GPT), so I couldn't get it to run correctly.
Round 2: Technical Interview This round didn't involve fixing the previous bug. Instead, I was asked many questions about my resume, as well as some standard interview questions. I had prepared for standard questions based on previous interview experiences shared online. A memorable question was how MySQL and DynamoDB implement pagination.
Round 3: AI Technical Discussion I didn't prepare for this round, learning from my previous experience of memorizing ML content. The questions were about LLM application-level issues, such as how to validate LLM output, how to solve hallucination problems, what problems LLMs and traditional NLP are suitable for, which NLP embeddings I've used, and how to modify the craft demo problem and explain the solution.
Round 4: Hiring Manager Because I didn't do well in the craft demo, the hiring manager didn't seem very interested in me. I asked why they scheduled an AI interview, and they said it was because they wanted to hire someone with LLM application experience since no one on the team had AI experience, except for the AI round interviewer, who had learned about it on their own.