Two AI coding rounds, one traditional system design, one AI-Enabled Design round.
AI coding 1: Max Unique Character Subset — given a list of words, find a subset of words where no two words in the subset share any repeated letters, and the subset covers the largest possible total number of unique letters.
AI coding 2: Kind of a mess. This round the interviewer opened by telling me he hadn't realized the question would overlap with my first round, so he swapped it on the spot for a Friend Recommendation System question. One thing to note: talk to the interviewer a lot while you code, and bring up anything you get stuck on right away instead of sitting on it.
SD: Design LeetCode. The main part was supporting a coding contest and the leaderboard. I talked about running the Code Judge in a sandbox/Docker. The deep dive covered how to scale up the code judge and the DB, and how to track the current state of the code judge.
AI-Enabled Design: Also a strange round. The interview prep materials said there would be AI assist, but the actual environment I got didn't have it. I asked the interviewer about it and he said he didn't know either, so we just moved on. The question was to design a code review agent. It felt like you could just design it by picking whichever agentic system pattern fit. Deep dive:
- Which parts need an LLM and which don't
- How to validate what the agent generates
- How the orchestrator and memory work
- What tools are needed
- How to calculate the cost of each LLM inference call
BQ: Two new questions were close to what was in the prep materials — AI-Driven Impact and Continuous AI Learning. The rest were standard BQ, asked pretty comprehensively — almost every BQ question came up.
Currently waiting on the HC (hiring committee).
Discussion
Loading comments…