Overall impression: this one was really hard.
First there was a 15-minute recruiter call — a quick chat about my current job, why I was looking, and visa/status questions. They said the interview process would be laid out in a follow-up email.
The phone interview was two rounds total, one hour each.
Round 1: coding (SQL and Python), plus a case study.
The SQL question was built around drivers and riders — calculating top orders, cancellation rate, and so on. There were a lot of filter conditions, and I had to be careful about when to use WHERE versus ON.
For Python there were supposed to be two questions. The one I got: my opponent and I are rolling dice, and my opponent is the house. If I win, they pay me n dollars; if they win or it's a tie, I pay them m dollars. Assuming the dice are fair, should I play this game?
Round 2: XP/Stats/Causal Inference.
This one was with someone from the airport team, and it was all about supply and demand for airport rides and reducing the cancellation rate there. They went incredibly deep and detailed. Airport is a hard place to test cleanly — you can't easily run a geo test or a switchback, DID can't guarantee it accounts for time-varying effects, and synthetic control is nice in theory but rarely used in practice. There's also network effects, and airports are just a special case in general: drivers there sit in an actual queue and have to wait their turn. When a driver or rider cancels an order, it hits the driver's experience hard — either they've waited forever and now have to leave the airport, or they have to loop all the way back around to rejoin the queue they were in. Riders at the airport are a special group too — whether they can find the right pickup location on their way there, whether they even pick the correct pickup location, whether they're a one-time user, whether it's a business trip or a personal trip — all of that matters. The whole case is very different from a normal city case. How would I measure driver and rider happiness/satisfaction separately, what angles would I consider, which of those are data-driven, and what other hypotheses do I have — even something from a psychology angle that isn't backed by data is fine to discuss.
Discussion
Loading comments…