Ray

CS & Math Stuff

© 2023. All rights reserved.

I interviewed for Google’s Tensorflow, Apple’s MLPT (Machine Learning Platform & Technology), Bytedance’s ad infrastructure, Databrick’s ML team, Citadel Securities as a quantitative research analyst, Hudson River Trading(HRT) as an algorithm engineer, and Jane Street’s research desk as SWE. I received offers from all of the companies except for Jane Street. Here’s my experience interviewing during COVID.

Disclaimer: I won’t be walking on the edge of leaking confidential information like an idiot(yes, I signed an NDA for all of these companies). Don’t expect to get any hints for your interviews.

The structure of this blog is inspired by my friend Han’s medium blogpost.

interviews

Table of Contents

Preparation

Working on machine learning infrastructure is 99% systems engineering and 1% machine learning. My experience on machine learning infrastructure teams has taught me this, and preparing for systems engineering topics was the right way to go. I did the following to prepare:

Algorithms

~50 leetcode hard questions. Some of them are DP, some are graph based, some of them are just NP-hard problems that are a pain to code(which is the point), and some include devising some clever data structure that supports a very specific access pattern. I gave myself roughly 40 minutes to solve these problems. ~15%(7 questions) of the time I couldn’t figure out the correct solution because time limit exceeded, memory limit exceeded, or I was just flat out wrong. I directly read the solutions and learned the tricks necessary to solve the type of problems moving forward. Don’t bother with medium or easy questions since hard questions often contain medium/easy tasks as subroutines, and these companies probably wouldn’t ask you easy leetcode questions anyways.

I wrote the solutions in either python and C++(sometimes both) and went back to polish my code for minor optimizations or readability improvements. For C++, I made sure I wasn’t using raw pointers unless appropriate and I was using C++17 (constexpr functions, std::array instead of raw arrays, smart pointers, template type deduction with lambdas, etc) features. The reason I wasn’t using C++20 was because the online coding platforms(like coderpad) likely use stable distributions of GCC and clang, which means some of the new features are in their experimental phase. I didn’t want to encounter a bug with concepts or std::ranges in the middle of the interview. (In fact, I found a bug with attributes recently in a new version of gcc)

I also spent a few days on problems elsewhere:

Systems Design

Working at Airbnb has made me pretty familiar with high level distributed systems design, but of course I worked only with a subset of them. I think Martin Kleppman’s book Designing Data Intensive Applications is a great read, but you’ll have to pick and choose which sections you want to go over as it’s a pretty dense book. If you don’t have time, maybe just try understanding how Kubernetes works with Marko Luksa’s Kubernetes in Action, which is a much easier read. You can then draw parallels with the distributed design for K8s against whatever systems design question the interviewer has for you.

Make sure you know some fundamental ideas about distributed systems like the map reduce paradigm, sharding, asynchronous and synchronous follower replicas, CAP theorem, etc. What you don’t want to do is read 3 sentences about each of the terms above and regurgitate it in your interviews. Interviewers have been doing this for a while, they know you don’t actually understand the concepts. Don’t be that guy.

Math Questions

These are only asked in finance firms. Honestly, these are just all over the place. I read this green book called A Practical Guide to Quantitative Finance Interviews by Xinfeng Zhou, but only doing a single problem in each section by myself. Hedge funds will quiz you on discrete math to probability theory to geometry to information theory to literally anything. My advice is if you’re a software engineer interviewing for a hybrid of finance and tech places, timebox yourself in this category.

I have not seen an interview question this cycle that was an exact question I’ve seen online or in books. Your mileage may vary.

The interview process

Interviewing and talking with all of these companies was a great experience, even with COVID in place. Obviously, as shelter-in-place continues, these companies are conducting virtual on-site interviews and trying to make this process as smooth as possible. Without getting into the specifics, I’ll outline some common things I’ve noticed during the process in the COVID era.

More interview rounds during COVID

The bolded text might scare you as a potential candidate, but don’t worry too much. The added questions aren’t testing you if you know how to implement a bloom filter or a fibonacci heap or something niche. They usually test on the coding abilities of the person and how well they’d actually ramp up in a novel, collaborative environment. This can manifest itself in multiple ways - live debugging session with a new codebase, reading documentation to work with new technology, or a collaborative brainstorming sesion for a hard(er) problem. If you’re a decent software engineer you shouldn’t worry about these as much.

Dealing with time zones

One of the biggest struggles I had during the interview process was adjusting my sleep schedule to wake up at 5-6AM to make sure I’m awake and on time for the interviews in New York/Chicago (I’m in California so this was a 3 hour gap). Usually, companies would fly you out the day-of or the day before the on-site. I’ve always felt tired after a plane flight and was able to get a good night’s rest before the interviews in the past. With COVID, everything is virtual and the companies expect you to interview at their hours.

Even with slowly adjusting my sleep schedule over a week or two I still had trouble with sleep. Personally, I get pretty nervous before an on-site and I’d need to feel adequately tired to get a good night’s rest instead of tossing and turning in bed. With the clock turned 3 hours back, I suddenly found myself not tired enough to sleep on time the night before the interview(even with a whole week of adjusting). This led to me consistently getting 6-7 hours of sleep instead of the 9 hours of sleep I usually get on game day, which really sucked.

Ultimately, I have no idea how much the sleep problem really affected my performance, but it was enough to shake my confidence going in.

NOTE: +1 to Citadel for proactively breaking my on-site over multiple days so I can have a sane sleep schedule for their interviews. This might depend on the specific team you’re interviewing with.

Which ones were the hardest?

This is subjective, and the question can be broken up into multiple components:

Once again, this breakdown is subjective. I obviously have a lot of experience interviewing with Silicon Valley companies so the novelty of questions from the finance companies added to the difficulty.

Making a decision

This was the hardest part for me. I spent two weeks suffering from analysis paralysis. I would wake up wanting to go to company X but wake up another day wanting to go to company Y. The tug of war between different recruiters stressed me out - I couldn’t sleep, I couldn’t eat, I couldn’t do anything during the days and I unknowingly stressed out those around me with the process (special shoutouts to Ben, Eric, Nishanth, Mickey and Kibeom for dealing with my BS). I used the following criteria to make my decision:

I made a weighted linear model consisting of these features and used that arbitrary numerical output to reduce my choice to between Citadel and HRT which were exactly equivalent in numerical value. In the end, I decided using my gut feeling and went with HRT.

The culture and the “small” things count

My decision was largely driven by the vibe of the people I talked to:

Aside from HRT, I’d like to thank Xinan and Xing for being amazing hiring managers who spent a lot of time talking to me and helping me throughout the process with invaluable advice. Their experience, honesty, relatability, and transparency made the decision so much harder to make (in their favor, of course). Because of them, I felt like I wasn’t just another data point in the interview pipeline, and that they were advocating for my success regardless of where I end up.

Although I felt like the decision making process was least impacted by COVID, if I were able to meet potential co-workers face-to-face it would’ve been clearer which place I would’ve liked to work at.

Conclusion

Interviewing during COVID is definitely a different experience. It was a lot of stress and I’m glad to be over with it. In the future, for my sanity, I would not go through the process with so many companies in different time zones at the same time. I’d like to thank my family & friends for supporting me through the entire process and cheering me on. I couldn’t have done it without you all :)