Did the Databricks coding round last week for a mid-level SWE role. Wanted to post the format because it's a bit different from what I expected based on older threads.
Format in 2026: two coding problems, 75 minutes, done over a video call with a shared coding environment (not a standalone OA platform). Interviewer was present the whole time, not silent, which I actually appreciated. They'd nudge if I seemed stuck.
Problem 1 was a medium-difficulty graph problem, BFS-based, probably LeetCode medium territory. Nothing tricky about the structure, but they wanted clean code and asked me to trace through a test case before I coded. I think they were watching for problem decomposition more than speed.
Problem 2 was harder. It was about interval merging with a twist involving priority. I'd call it a hard medium or easy-ish hard. I finished a working solution with about 10 minutes left, and spent that time cleaning up and adding a quick complexity discussion.
They asked me to state time and space complexity for both before moving on. Standard.
What surprised me: they didn't ask any Spark/Databricks-specific coding questions. No distributed systems coding, no SQL in the coding round. That stuff comes up in other rounds apparently.
Stumbling points I noticed: If you don't talk while you think, it gets awkward. They wanted active communication. They asked "is there a more efficient approach" after my first solution to problem 1, which was O(n log n). I found an O(n) version on the spot but honestly that felt lucky.
Overall: LeetCode medium is the right prep level. A couple hard-mediums in there. Not consistently hard. But the live format with someone watching changes the stakes a bit if you're not used to it.