just finished the Atlassian online assessment and then the coding round in the actual loop. different from what i expected so sharing the full picture.
online assessment (OA): this was on HackerRank. two coding problems, 90 minutes total. difficulty felt like leetcode medium. one problem was string manipulation (think parsing a structured log format), the other was graph traversal. nothing DP-heavy, no segment trees. if you can reliably solve mediums under time pressure you should be fine.
the actual coding round (in-loop): this is where it gets more interesting. one hour, one problem, one interviewer. for me: given a stream of events, group them by session (session = events within 30 minutes of each other for the same user). classic sliding window / interval merging type problem.
they cared a lot about: starting with a brute-force approach and narrating your thinking before jumping to the optimized version. code quality. they literally said "write production-quality code, not pseudocode." edge cases, clear variable names, a little bit of error handling. testing. after i wrote the solution i walked through test cases and they asked what else i'd test before shipping this.
i'd say it was medium difficulty but the bar is on code clarity as much as algorithmic cleverness. i'm a frontend dev and a lot of my work involves event-stream processing so this was actually comfortable, but the testing emphasis surprised me.
python vs. java vs. JS: anecdotally: the interviewer seemed fine with whatever language. i used python. friend who went through it a month ago used java and had no issues.
bottom line: prep mediums with focus on interval problems, sliding window, and string parsing. do not skip writing clean code because the interviewer is specifically watching for that.