Did this a few weeks ago while employed. Here's what the Ramp coding rounds actually look like since the information out there is weirdly sparse.
There's no traditional OA first. They skip the automated HackerRank-style screening. First coding touchpoint is a live interview with an engineer, about 60-75 minutes.
Format: One to two problems per session. If they give you one problem it's usually more complex. The second problem (when they give one) tends to be shorter and tests something specific like a data structure or string manipulation.
Difficulty: Mid-to-hard LeetCode equivalent. I'd call most of them "medium" in LC terms but they add context that makes them harder: the problem is framed as something a real fintech system would do. One of mine involved transaction categorization logic that required careful handling of edge cases in nested category hierarchies. Not a standard tree problem, but it drew on tree traversal skills.
Expect follow-ups on time/space complexity. They'll also sometimes ask "how would you change this if the dataset was 10x?" which is a distributed systems angle in disguise.
Language: your choice. I did Python, which was fine.
What tripped me up: I spent too long trying to optimize upfront. They care that you get a working solution first. Optimization is a follow-up discussion, not an upfront requirement. I learned that in round 2 the hard way.
Overall the bar felt real but not unnecessarily brutal. If you've done honest prep (not just LC grinding, actually understanding the reasoning) you'll be fine. They're not trying to catch you out.