Tower Research Capital Data Scientist Interview Guide

I went through a pretty long process with Tower Research Capital, and the most memorable part was how much it mixed theory with practical ML discussion. In total there were 10 rounds, though in my case it also included a Hackerrank take-home that was a 3-hour notebook-style test with multiple research problems. The instructions were pretty vague, and one of the coding prompts even had an incorrect description, so it felt a bit under-specified for the amount of work expected. It also came across more like something aimed at a new grad than an experienced candidate.

The live interviews were mostly technical and fairly repetitive across rounds: linear regression, probability, coding, and general ML fundamentals. A few of the regression questions were genuinely hard, not just standard textbook stuff. One that stood out was a correlation puzzle: given Corr(X,Y) > 0, Corr(Y,Z) = 0, and Corr(X,Z) > 0, they asked which coefficient would be bigger between Y ~ X and Y ~ X, Z. I also got asked about L1 vs L2 regularization, lasso versus ridge, and the difference between random forest and gradient boosting. There was one brain teaser style question too, including a banana-transport puzzle. The final round was onsite, and they seemed to care a lot about whether I had real hands-on ML experience. They asked very concrete questions about what features I had built, which dataset I used, which models I chose, and how I converted predictions into trades. I heard back fairly quickly after most rounds, but the final round took longer. In the end they said I didn’t have enough practical ML experience for the role, so I didn’t get an offer. My main takeaway is to be ready for both tricky regression/statistics questions and very specific questions about how your past ML work actually translated into production decisions.