Ace The Data Science Interview

It’s a lot, and can feel overwhelming. First things first — ask recruiters what you need to know, check this via Glassdoor, or better narrow down your search and only apply to most relevant titles. For example… System Design might be more for Data Engineer or ML Eng roles… not DS or Data Analyst. Similarly, ML Eng might not be tested on SQL but more LeetCode style question. This narrows down the space of things to know.

But to get to the core of the question; balance. I have no idea, it’s deeply personal. I struggle to balance cooking healthy, working out, with my career stuff. Others have their own balancing act issues, balancing job search stuff with family friends and full time jobs. Ultimately comes to prioritization — there’s no easy answer!

Nick Singh yes the 2nd part of balancing job along with other commitments along with family how do you do that? Would love to know that also 🙂
Also to add a lot of them are also asking these product manager type of questions like how to improve Google Maps etc

echoing what Nick said - focus on the particular job you want and what those job descriptions encompass. Product analytics roles will heavily focus on SQL/Product sense vs. ML eng which is coding & ML for example. Also depending on the company and role you will want either more breadth (FB product DS) or depth (Netflix ML research) so you should allocate your studying time accordingly

spaced repetition. I’ve been exposed to it, multiple times, over multiple years, that even if I don’t remember every detail I know what I can look up, or can easily refresh myself. No other way around it!

Also it helps to practice the fundamentals - ML (like many technical topics) is not easy for rote memorization, and being able to pattern match among different use cases & fields is nice

Coding questions often asked are too hard. When was the last time a DS needed to know about a tree structure or used a Linked List. Come to think of it… do even SWEs use Linked Lists? I think those leetcode style questions are already sorta dumb for evaluating SWE talent, and lazy thinking ported it over to DS.

What Nick said also applies to hard technical questions in ML/Stats/Prob land for ML research / ML eng and quant roles (in finance) although it’s more relevant since for deeply technical roles you should know the fundamentals well (for example if you are doing ML research you should understand the math behind the algorithms)

An early filtering round. Quick phone call about projects/background, throw in 1-2 stat/ML theory questions, and have them write a simple SQL query. Maybe of load some of this to a recruiter if possible. Then do 2 more technical rounds. First technical round is something more academic/textbook… aka a data manipulation/cleaning exercise, some more theoretical questions (what’s your favorite modeling technique and why… probe if they just call sci-kit learn or really know what they are doing and understand the fundamentals). … and then something more open-ended / how would you approach this real-world problem we are dealing with. Do they have business sense. Then a final behavioral interview to understand what this person is done, how fast they are growing, and whether they actually are interested in the job.

Depends on the role - product analytics: SQL & product sense, ML eng: coding & system design & ML for example, and also the company (higher % of more senior ppl at Netflix than Facebook). More variance for startups since the hiring process is less defined. But yes what Nick outlined is generally good for a DS process

1. one way I have found effective is to do mock interviews with friends or others - when you can explain deeply technical concepts in a succinct manner then you understand them quite well. also one tip for me personally is not memorizing any specific equations - if you focus just on rote textbook memorization, then when thrown a curveball, just like an ML algorithm you’ll have overfit your knowledge. 2) doing projects is super helpful here, to demonstrate that you are proactive, driven, and passionate about using data science for XYZ

Thank you for the reply and i will follow these advices to improve myself. 🙂 🙏

1. depends on your background (they will ask about your research experience if you have any) and then depending on research or eng will be more or less theoretical, 2) no since otherwise it’s a chicken & egg problem - that being said, the interviews are there for them to gain confidence that you understand the models well enough theoretically and also have some experience (whether through projects or research etc) to be a good fit

Thank you Kevin for your response. I really appreciate it 🙂

Might be some personal bias here but things I look for are: 1) motivation for joining company/team specifically, 2) where they see themselves in X years, 3) what gets them excited to do meaningful work everyday, 4) their ability to be a team player and questions that are in those areas may range from something like “tell me about a time you handled a conflict with a boss” (for 4) to “I saw you had XYZ on your resume as projects, how did you come about those?” (for 3)

Again personal bias here but generally along the lines of what the company is looking for and how to grow within the company - examples: 1) what does the day-to-day look like? 2) how does the role change over time? where are the biggest areas of growth? 3) what is the company culture like? what about trade-offs (for example, speed of shipping new products vs. having highest quality)?

Absolutely! STAR is for behavioral questions.. I think of it as structured story-telling. And DS & ML roles def ask these types of questions!

+1 on what Nick mentioned (he’s a verified pro with the method!)

Depends on company and role again but those topics should only be asked if you have relevant experience/projects in real-life large-scale model deployment and or if the job description mentions MLOps specifically

Practice makes perfect (generally) - just like an ML algorithm you get better with more data (interviews) 🙂 - also need to know when to focus on breadth vs. depth based on the roles

When most job descriptions have that job req, it (usually) isn’t a hard requirement, they just need a way to filter candidates. There are obviously exceptions, especially for deeply technical roles requiring specific knowledge (say ML research). My 2 cents here would be to try and do projects in the space to demonstrate drive and competency

Thank you Kevin for the answer 🙂

Definitely depends on company and role - from the behavioral side it’s a sense of motivation and drive and cultural fit, from the technical side it varies a ton but is generally related to either breadth or depth among a variety of topics. For example, FB product DS is about breadth and knowing SQL/product sense etc vs. Wall St quant is about stats/prob/coding/ML generally. Sorry it doesn’t answer the question but that is my observation, there is no silver bullet 🙂

Can be picked up on the job for sure

Generally those job descriptions are looking for lots of depth (specific knowledge) so good to focus on those areas. Projects are great as you mentioned, keeping up with state-of-the-art techniques also (if ML), and just generally being able to demonstrate that even if you have no direct job experience in the space, you can and will pick it up quickly

Nice to do both, with a focus on the fundamentals (i.e. not memorizing and equations or specific syntax)

great question! you should mention it, if you could talk intelligently about it. If you lack work experience… and just need to fill it up.. then def add more things. But don’t feel the need to over-stuff with random things… sometimes if something isn’t very special … but other things on the resume are… adding a simple neutral project detracts from the good stuf

+1 to what Nick said - for me personally I only include it if I feel ready to be (potentially) grilled on the topic

Depends on the person and how much time spent per day - generally a few months of full time effort should be good enough to be quite proficient in any area although the more deeply technical the topic the harder. So should be an okay timeline roughly speaking

Not high signal unless you’ve won competitions imo

Focus on the major ones, but generally most roles look for similar things. So one way to get a sense of what’s important is just to look at the same type of role… and across multiple companies you’ll get a sense of it. For example, for most Product Data Science roles you’ll see mentions of Exploratory Data Analysis, SQL, A/B Testing, and “supporting product/business teams” == biz acumen and product sense.

+1 to what Nick said, for entry level it’s more about the fundamentals than anything too specific so be sure to brush up based on the relevant area (Product analytics is a lot of SQL & product sense for example, whereas ML eng is coding & systems design & ML)

Definitely depends on the person - with so many topics it can feel like a firehose (I personally like doing brief deep dives but then rotating among topics). If you get asked those topics you should just admit you don’t know / don’t have any experience in the space - in general interviewers shouldn’t press too hard for this as long as you don’t put random buzzwords on your resume that you can’t explain thoughtfully

Definitely side projects, and any research if relevant (for example, lots of ML hiring managers love research). Also helps to stay up to date with state-of-the-art and whats going on in open source world to demonstrate passion in the area

Yes, that’s what I had thought, and good point about the research component.

Good to know both and understanding tradeoffs if asked about them - still good to start with the basics, but if you’re applying to jobs in a specific space (say CV at self driving co’s) then it’s good to be able to keep up with and explain state-of-the-art techniques

Thanks!

Definitely - will let Nick chime in on the behavioral side but for the technical side some things include: 1) not assuming you know more than your interviewer especially if it’s specific in the context of the job, 2) pretending to know some technical topic then getting grilled on it and not being able to explain anything, 3) talking/coding/whiteboarding for way too long on a topic without checking in with the interviewer. Have been through all 3 personally :)

Will let Nick chime in on the behavioral side - on the technical side: 1) figure out whether you’re optimizing for breadth and depth since there’s an infinite among of knowledge in the spaces that make up “data science” 2) the more work experience you have, the more emphasis that will play in interviews (generally) and if you have less experience then projects are great, 3) fundamentals matter more than rote memorization especially for very technical topics like Stats & ML

Thanks Kevin definitely
useful summary.

Will let Nick chime in here - it should be suitable for all levels (we have problems ranging from “easy” to “hard” and across many companies. That being said, it has a lot of refreshers on topics, chooses specific topics, etc. and therefore is by no means a silver bullet for every person and use case

There’s a lot of variance on takehomes.. will let Nick add here but generally speaking my personal view is that it’s there to filter for candidates who truly want to work at the company since it’s generally not an easy task. So would only do it if you really want to work at the company, and would always timebox it (can always spend a long time but I personally prefer a quicker, mostly complete project than a completely perfect one since the marginal utility is lower for the latter imo)

Will let Nick add on here but for me anything demonstrating curiosity for the role and the interviewer themselves, examples: 1) what brought you to XYZ company? 2) how do you see your role changing over time? 3) can you describe your day-to-day?

there’s a good amount online about various companies so it’s just about doing your homework - yes generally new grads may have a worse grasp of business sense but that may change with technology adoption trends and a growing awareness

depends on role and company but data analyst / product data scientist is usually focused on SQL / product sense whereas MLE (and NLP & CV engineer) is more coding / systems design / ML

As a side note, your book was incredibly helpful for me in preparing for DS interviews - and yesterday I actually got my first job offer! :thank_you:

Congrats Gur Hevroni! Glad to know it helped 🙂 Regarding your questions: 1) its a numbers game and depending on the person it’s useful / not useful to try and analyze that metadata for takeaways (you can even try to apply some data science on the process if you interview at enough places haha), 2) doing mock interviews with others and feeling ready to be able to explain projects and various concepts are good indicators

Thanks, Kevin! These are great tips!

Will let Nick elaborate here, but my 2 cents is that it’s important to just choose an area you’re truly interested in and go from there - DS/ML is a toolkit and can be applied to virtually every field. The importance in getting initial interviews and in the process itself varies with role/company but I find that for the former, it generally has a decent weight the less direct job experience you have, and for the latter it matters a lot in technical roles (ML especially)

Thanks Kevin

The business sense is super useful and helpful for product analytics / product DS roles so high generally

1) I’ve looked at some of MIT OCW for technical topics and for ML there’s a lot out (Andrew Ng’s material for example) there so will need to come back to this. 2) they generally act as a technical PM so basically they interface with ML engineers/researchers etc. and hence very similar to regular PM except that they need to understand and speak the technical jargon