Went through two rounds of their online assessment plus two live coding phone screens recently. Here's the honest breakdown because I wish someone had told me what to actually prepare for.
The OA
Two coding problems, 90 minutes total. The platform was HackerRank. You get to choose your language. I did Python and had no issues with the environment.
Difficulty: one medium, one that I'd call medium-hard. The medium was a graph traversal problem with a twist (edge weights with conditional behavior). The harder one involved some number theory that caught me off guard. If you've only done the 'easy and medium Leetcode grind' you might hit a wall on the harder problem.
There's no proctoring camera on the OA, but don't assume that means it doesn't matter. I was told the solutions get reviewed by an engineer, not just auto-scored. Readability and correctness of your solution structure apparently matters.
Live coding phone screens
These were more open-ended than the OA. One interviewer asked a problem I'd classify as Leetcode hard (interval scheduling with a resource constraint I hadn't seen before). The other was a graph problem that was medium but with a follow-up that pushed into hard territory.
Both interviewers were fine with me thinking out loud and asking clarifying questions. One explicitly said 'take a minute before you write anything.'
What I'd prepare: Leetcode mediums until they feel easy, then start adding hards from the graph, DP, and intervals categories. The string manipulation and array sliding-window stuff is less likely to appear at the senior level. Also review time and space complexity formally. They asked me to prove my solution's time complexity in one round, not just state it.
Not a leetcode grind factory in the worst sense, but you need to be solid. It's quant finance. They hire people who can actually think about algorithmic complexity.