The recruiter got back to me within a day, really fast.
First round, coding — 3 levels, 2 of which were the bar to clear. You write a schemaless database (an in-memory SQL), but you're given each column's type and name. Some rows might be missing certain columns. The first level is figuring out what data structure to use, then basically being able to store and read. The second level does filter and projection — specifically, you use an API to do select and filter, then project into another column name. Time was very tight. There were a lot of edge cases to watch for on the details, like a column not existing, or a column's type being different from expected.
Second round, hiring manager — about 45 minutes mostly talking about projects I'd worked on, traditional behavioral STAR questions. Topics included product challenges, complexity, and team communication.
Third round, onsite — 4 rounds total, could be split across 2 days, and if you didn't pass the first day it ended right there. I asked to have everything on one day, one hour each. They were:
Coding — similar style to the phone screen, 3 levels. The interviewer said you had to finish 2, and the 3rd was a bonus. Level 1 was calculating a loan's principal and fee balance, with a few tests given to try against. There were a large number of events, and for each event you had to separately figure out how much went to principal versus fee — fee gets paid off first, then whatever's left goes toward principal. Level 2 dealt with events that could arrive with inconsistent timing: in level 1, principal and fee came in as a single event; in level 2, principal and fee were split into two separate events, and the problem required that you wait until you'd received both the principal and fee events for the same time period before calculating, then continue on to the following events. Level 3 was implementing logging using batching.
System design — design a scheduled payment system for a mortgage, arranging payments and splitting payments across different investors.
AI coding — design a fairly standard CRUD FastAPI server plus a rate limiter and auth (JWT), with parts that needed a thread pool and retries for handling remote data calls. Time was a bit tight. It was the interviewer's first time running this round, so there were some delays. I think it was mostly meant to be open-ended, seeing what you think a modern API design should include — whatever you mentioned and confirmed on, you'd then have AI write it out.
Project deep dive — required using 3-4 PowerPoint slides plus Excalidraw to draw out the system.
Discussion
Loading comments…