First recruiter reached out to me and we chatted, then general chat with hiring manager which went very well. Next was technical phone screen which was challenging and very stats oriented. Apparently, there were some "concerns" but they still wanted to move forward and give me an exam. After this I got a ridiculously long take home exam, I spent literally 10-12 hours working intensely on it, but easily could have spent much more and I am somebody that can work quickly. The 3 sections were all reasonable but just was a lot to ask. After turning it in, I heard back from the recruiter that said I scored very well on it (apparently they send the test to a separate Data Scientist outside the team you interview with to grade it). I was happy to hear this and expected to be called in for the final round in-person interviews. But then the recruiter told me that the hiring manager wanted to chat with me again on the phone. I agreed and expected a chat about the test I turned in or perhaps just more background on what the role would be and whether it made sense for me to go in for the in-person final round. Instead, after telling me he hadn't seen my exam nor the score for it and proceeded to give me ANOTHER technical phone screen which I was NOT expecting. I was annoyed but still tried to answer the questions but it did not go smoothly. I then heard back 2 weeks later that they did not want to move forward. Think very carefully about whether you want to invest the time in doing the take home exam. Overall, a terrible experience.

Recruiter call to just check if you get how a two sided marketplace actually works and if you align with their core values. After that was a 45 minute live screen that was mostly advanced SQL window functions and some basic metric diagnostic questions. The onsite was a 5 round loop covering product sense, stats and experimentation, applied modeling, data processing, and a behavioral bar raiser. The stats and experimentation round is the real filter i think. You cannot just suggest a standard A/B test for a new feature. They really push you on network effects and driver cannibalization, so you have to know switchback experiments and synthetic controls inside out. Product sense was basically a deep dive into root cause analysis, like being asked to figure out why rider cancellations suddenly spiked in a specific city and walking through the exact metrics you would pull. The modeling round was less about writing math on a whiteboard and more about how you handle imbalanced data and pick the right tradeoffs for putting things in production. The bar raiser chat is pretty intense. They will dig deep into your past projects to see if you actually drove the business impact or just wrote the queries, and they care a lot about how you push back on product managers. For prep, do not just grind leetcode database questions. Practice structuring ambiguous product metrics and read up on their engineering blog. Doing a mock on prepfully with an Uber DS helped me to catch my blind spots with the switchback experiment stuff and get a reality check before the actual loop. Tough process overall but really engaging.