Applied to a couple of Vanguard SWE roles this spring and went through their online assessment for one of them, plus a technical phone screen for the other. Sharing what I saw because the format is different than the big tech OA experience.
Online assessment (OA): Hosted on HackerRank. Two coding problems, 90 minutes.
Problem 1 was a graph traversal problem. Not the hardest variant, more like a medium on LeetCode. BFS or DFS both worked. I did BFS, got full credit.
Problem 2 was finance-flavored: given a list of fund transactions, compute some kind of running balance or rolling window metric. Felt like an array/sliding window problem dressed up in financial terms. If you panic at domain terminology, breathe through it and translate it to the underlying data structure problem.
Time limit was generous. I finished in about 55 minutes and had time to check edge cases.
No SQL in the OA I took (this may vary by role). The job description for the data engineering track probably gets a SQL component, I'd guess.
Technical phone screen (different role): Live coding in a shared editor, 45 minutes with a Vanguard engineer. One problem, medium difficulty, string manipulation plus some hash map usage. They also spent about 15 minutes asking me to walk through my past projects. Not a grilling, more like genuine curiosity about what I'd built.
Overall difficulty: Honestly below what I expected going in. Not a surprise given they're not competing with Google for engineers. But don't go in underprepared. Mediums are still mediums. And if you haven't touched graphs or sliding windows in a while, refresh those.
One thing they specifically called out in the debrief: they care a lot about code readability and error handling, more than raw cleverness. Writing clean, well-commented code with thoughtful edge case handling seems to score well here.