Did the HF coding rounds twice now: once for a role I didn't get a few months ago, once for a different team more recently (got that one). So I have more data points than I'd like.
Online assessment (round 1 or 2 depending on the team): Not all teams use one. My first loop had a timed OA through HackerRank, two questions, 90 minutes. Difficulty: one easy, one medium-hard. The medium-hard was tree-related. My second loop skipped the OA and went straight to a recruiter screen then a live coding session. So it varies.
Live coding: CoderPad, share-screen optional but they don't require it. You can use any language. Python gets the most comfortable response from interviewers in my experience, but they didn't penalize me for using Go in the second loop.
Difficulty breakdown from my experience: First loop: medium, medium-hard. I failed on time for the second question in round 1. Second loop: two mediums, both completeable in the time given if you don't overthink it.
Style of the problems: Not pure Leetcode grinding problems. They felt more practical: "parse this structure and return X," "given this dataset find Y efficiently." Still algorithmic but grounded. One problem in my second loop involved working with a JSON-like structure, which felt very HF-flavored.
The interviewers in live rounds actually talk. One of them walked me through a hint when I was stuck, and didn't tank my result for needing it. That's different from some places where the interviewer just stares at you.
What I'd prep: BFS/DFS fluent Tree traversal fluent String manipulation Hash maps for everything Light graph stuff
You don't need to prep DP deeply. It didn't show up in either of my loops.