2.1.3 What signals companies look for in candidates

Accurately evaluating candidates is very challenging. First, you can only evaluate something as well as your evaluators allow. Companies can only evaluate a candidate to the extent of the interviewers’ knowledge. If your interviewer has a shallow understanding of X, they won’t be able to evaluate your in-depth understanding of X. Many companies, including those who claim to be ML companies, don’t already have a strong in-house ML team to act as good evaluators35. Second, even strong in-house teams don’t always mean strong evaluators. Therefore, companies have to rely on signals to help them predict whether a candidate would be a good fit.

As you might have already suspected, pedigrees make for strong signals. It’s not a coincidence that companies like to advertise how many ex-Googlers or ex-Facebookers they have on the payroll. If you’ve worked as a full-time ML engineer at Google, you must have passed its ML interviews and learned good engineering practices from Google.

On resumes, college names matter but not much. Their importance is inversely proportional to seniority. If someone, with all the privileges of an elite education, still has no interesting past projects to put on their resume, the fancy college name might even hurt.

However, going to a popular engineering school has several benefits. First, given two equally mediocre resumes, one from MIT, the other from a college nobody has ever heard of, the recruiters might be more inclined to give the one from MIT a call. Second, popular engineering colleges give you access to recruiters who hire from campus events. Third, you’ll likely have classmates at big companies who can refer you.

If you’re a recent graduate, your college name might matter less than your GPA, which shows your dedication during your studies. Still, your GPA doesn’t matter as much if you have other things to show. I’ve had only one employer asking for my GPA, and it was after I’d got the offer so that they could put it in their database.

The strongest signal is past experience, especially experience similar to the job you’re applying for. The experience can be work done at your previous jobs, projects you do independently, or competitions you enter. If you’ve placed highly in Kaggle competitions, made significant contributions to open-source projects, presented papers at top-tier conferences, written in-depth technical blog posts, self-published books, or done any interesting side projects, you should put them online and highlight them in your resume. There are so many things you can do to signal to people that you’re proactive, capable, and willing to work hard.

When I asked on Twitter which signal is most important when screening for ML engineering roles, more than 50% of the 2458 respondents picked GitHub/Kaggle. 22% chose previous employers, 15% picked referrals, and only 10% thought school names were the most important signal. There’s a lot of noise in this survey because it’s possible that people picked the signal they wanted to be important instead of signals that were actually important.

What signals companies look for in ML roles

⚠ The free project bias ⚠
In 2013, Chris Anderson, the author of The Long Tail, tweeted about the advice he received about hiring software developers: “reject anyone who doesn't have a GitHub profile (the more active the better).”

Even though GitHub/Kaggle in particular and past projects, in general, seem meritocratic, we have to be mindful of the candidates’ circumstances when looking at them. Not everyone can afford to contribute to open-source projects or enter Kaggle competitions. If we place too much importance on voluntary activities, we accidentally punish candidates from less privileged backgrounds -- those who work long hours, have too many responsibilities at home or face online harassment for who they are.

One group that suffers if hiring decisions are made based on open-source contributions is women. According to a 2016 research by the National Center for Women & Information Technology, the percentages of women in various software engineering occupations are 21% (Computer Programmers), 18% (Software Developers), and 34% (Web Developers).36 Yet, according to reports on Toptal37 and Wired38, only 3-5% of open-source contributors are women.

Some hiring managers are aware of this privilege bias. Jeremy Howard, an ex-president of Kaggle and co-founder of fast.ai, responded to my survey on Twitter that he evaluates candidates’ achievements with respect to their backgrounds: “I look for people that have achieved an unusually high level of capability despite limited opportunities or significant constraints. It's been the best hiring signal over many years and companies for me.39

🌳 Tip: Sell yourself. Highlight your qualities 🌳
The hiring process of most tech companies, including and especially the biggest ones, is far from perfect. It's riddled with biases and loopholes. Yet, it's still being used because of legacy and bureaucracy. Until a better process comes along, the best that candidates can do is to understand the signals employers look for and maximize our visibility. On average, recruiters spend only 7.4 seconds on a resume. If you're a great ML engineer but can't signal to recruiters that you're amazing in those 7.4 seconds, you're out.

35: 40% of AI startups in Europe don't really use AI according to “The State of AI: Divergence 2018” report by MMC Ventures.

36: https://wpassets.ncwit.org/wp-content/uploads/2021/05/13193304/ncwit_women-in-it_2016-full-report_final-web06012016.pdf

37: https://www.toptal.com/open-source/is-open-source-open-to-women

38: https://www.wired.com/2017/06/diversity-open-source-even-worse-tech-overall/

This book was created by Chip Huyen with the help of wonderful friends. For feedback, errata, and suggestions, the author can be reached here. Copyright ©2021 Chip Huyen.

results matching ""

    No results matching ""