Databricks AI platform team, Staff level. I reached out to the hiring manager directly on my own and got the interview that way — it went straight to the VO (virtual onsite), skipping the phone screen. The initial loop was 5 rounds, then they added 2 more rounds afterward, and I still got rejected. I think I answered well. This is just how Databricks is — they don't really seem to have much sincerity about hiring.
All of it was previously-asked interview questions:
- Coding — an RLE/BP-style question. A bit convoluted. I talked through the logic with the interviewer as I wrote it, and once I gave a few examples it clicked.
- SD1 (high level) — bookstore. They drilled into fine details, like why the HTTP API design should be POST instead of GET, and how to batch the book price fetches.
- SD2 (low level) — log writer. They asked about WAL log batch writes.
- Project deep dive — the interviewer started interrupting from the first minute. The questions were good and deep, but the answers were actually in the part I hadn't gotten to yet. He was honestly a bit too impatient — getting interrupted that many times meant I never finished. Still, it felt okay overall, he was just a bit impatient.
- BQ (behavioral) — this one was with the HM. The HM was nice and clearly wanted to let me through. Behavioral questions are just something you get smoother at with practice — by this point, other companies' interviews felt like reciting lines with my eyes closed.
Then they told me they wanted to add more rounds. The interviewers for the extra rounds clearly hadn't been briefed properly.
- SD1 (high level) — bookstore again. I told him I'd already done this one. He asked why I'd need another SD round if I'd already done one — I said it was an add-on round. He reluctantly switched it to WAL log batch write on the spot (even though I'd already done that one too, I didn't feel like saying so again). I wrote it out once more, and it ended early.
- SD2 (low level) — implement a KV store. I finished the basic implementation and locking, and then batch write came up again. (Batch write is just the thing that's always top of mind for Databricks interviewers — once you fully understand it, you're set, because it's guaranteed to come up.)
The next day they told me I was rejected. The people who interviewed me were all senior staff with really strong backgrounds. I don't think I choked, though — I'd been asked about batch write so many times by then that I could write it with my eyes closed. I think the bar is high, but I also think I already met the bar. I just think Databricks' interview evaluation is really unreasonable.
Seven rounds total, back and forth — I felt strung along. Still, the intense prep built up a lot of interview experience for me going forward.
I'm also sharing the system design prep notes and high-level practice materials I put together for this.
Discussion
Loading comments…