Hiring Data Scientists with Roy Keyes

Listen

Transcript

[00:00] Tim: Welcome to episode 6 of the Into the Hopper podcast, an occasional podcast about machine learning, data science, software engineering, and more. On this episode, I have my friend, Dr. Roy Keyes, to talk about his new book, Hiring Data Scientists and Machine Learning Engineers. After a career in physics, Roy pivoted to a career in data science, and after years of being involved with the data science hiring process, Roy decided to put his pen to paper to share his lessons with others. Welcome, Roy.

[00:28] Roy: Hi, Tim. Thanks for having me.

[00:30] Tim: Would you like to add anything more about an introduction to yourself?

[00:33] Roy: Sure. I mean, I think you covered most of the high points, but I guess I’ll say my background is in physics. I studied computational physics, focusing on radiation physics for medical applications like cancer therapy with the radiation. And then about almost 10 years ago at this point, switched over to data science. And since then, I’ve worked mostly in the startup world, tech startups, kind of starting out with consulting, and then I went and spent a number of years at different startups working directly, and then most recently been doing some consulting again. And during that time, went from being an individual contributor to being a manager and spent a lot of time on hiring, yes.

[01:19] Tim: Yeah. So what was the— I mean, that’s a little bit of the background, but what motivated you to put together this book?

[01:26] Roy: Well, so last year in 2020, I was at a startup that not too long after the pandemic hit the US, that startup ended up laying off about half of the company, including me and most of the data science organization. And so after that, I started doing some consulting stuff, but at one point I wanted to look into the possibility of pursuing some of my own sort of business ideas. So I guess potentially doing my own startup. And there are several areas that I’m interested in, but one of them is around hiring, just because especially in my most recent role, I had a lot of issues around trying to scale the process of hiring and ended up evaluating a ton of tools to try to help out with hiring and skills assessment and things like that. And I was not very happy with the tools that were available on the market.

So I was exploring some ideas like that, and I ended up interviewing a ton of data science and machine learning hiring managers. And at the same time, I was doing some consulting stuff for a number of companies and basically ended up having the same conversation about hiring over and over. What kinds of roles should we hire for? How should we structure our team? How do we actually assess these candidates and attract them and things like that? So after a while, I actually kind of landed on not doing the startup because I couldn’t really find an idea that I thought was the right one. But at the same time, I had been having all these conversations and thinking about all this stuff that kept coming up over and over again.

And I was actually planning on doing a series of blog posts around hiring, but then one of the people that we both know, Joel Grus, I saw that he had published his most recent book and he had done that as a self-published book using a platform called Leanpub. And I naively thought, hey, I could probably just put out a book instead of some blog posts and it’ll only take me a couple of months, right? So, you know, here we are months and months and months later, and today is the 29th of June, and it’s— oh, sorry, it’s the 30th of June. Spent too much time working on getting everything perfect yesterday, but the 30th of June, and just launched the book.

[03:55] Tim: Cool. So, maybe before we dive in, it would be helpful to give something of an overview of what does this book actually cover, and for folks who are interested, which hopefully is everyone who’s listening, you can go on to the site, which is dshiring.com. Is that correct?

[04:14] Roy: Yes, DS as in data science, dshiring.com will get you there.

[04:18] Tim: And you can see the table of contents and there’s some sample chapters you can take a look at, but maybe you can give a little overview of what you tried to cover in the book.

[04:27] Roy: The book has a broad audience, I would say. It is aimed at, kind of on the one side of the spectrum is the data scientist or machine learning person who has never hired anyone before, but they’re suddenly in a position where they need to hire more people for their team. So this book covers a lot of just the process of hiring in general, but it’s also aimed at people who have experience hiring technical people, maybe software engineers or whoever, but they’ve never hired for these specific roles, and then in between. And I think it will also be of interest to people who have done this kind of hiring but just want to get additional ideas to help them improve their hiring process. So, there’s a lot of stuff there.

It starts out by talking about what is data science and machine learning, machine learning engineering, how does that fit into different businesses, and then asking you to kind of step back and say, what are we really trying to do and does it make sense to have these roles? And then what are the roles based on what we want to accomplish and what tasks are associated with those goals we have? And then kind of thinking also about some of the skills required to accomplish those tasks. And from there, really kind of narrowing down what roles do we really need? And then there’s a lot of discussion about what these roles are, which that’s probably one of the things that there are a billion definitions out there and everybody likes to— kind of argue about it and have their own opinion.

And I make a futile attempt to try to clarify those things a little bit to help people, you know, land on the roles that they really need. So, you know, once you have those in mind, then it kind of goes through and asks you questions about what your process should look like. You know, what are your resources available? What are the constraints placed on you by budget and time and people available to work on this? And so it sort of steps you through how to optimize this process relative to also the market. So if you’re hiring for especially junior-level roles, you’re going to get flooded with applicants. So you really have to, from day one, be thinking about how can I build a process that can handle that volume of applicants?

And then talk specifically about doing skills assessments and all your different options for doing that and doing interviews and how you structure them and then how you make decisions. And so it’s really the full spectrum there. And then finally, the last chapter is about setting up these new hires for success. What should you be doing there? And then on top of all of that, there are right now 5 interviews with other data science and machine learning hiring managers from a bunch of the different tech companies, and they offer a much broader perspective.

[07:26] Tim: Yeah, I thought those interviews were a really neat addition. I didn’t realize you had been planning to add those, and I think that’s an extremely helpful idea in a book like this to get some outside perspective. Did you find that— I guess, well, did you ask the people that you interviewed the same set of questions, or is it distinct questions for different people?

[07:45] Roy: So, I asked them basically the same questions. One part was just setting up the context of who they were and what they had been doing and who they’d been hiring. And then mostly I asked them, you know, what are the challenges you’ve seen? What are the changes you’ve seen in hiring? And then some more specific questions specific to the situation that they’re in. But I also was really glad that I did those interviews. I kind of did them for 2 reasons. I mean, coming out of all the sort of interviews I did when I was looking at startup ideas, I realized that my specific experience hiring these kinds of roles was certainly not encompassing all of the experience out there by a long shot.

But specifically, I had been in these small to medium-sized startups, and had a lot of resource constraints and kind of specific scenarios. But talking to people, especially from some bigger tech companies, you realize that the kind of challenges they face are very different than the ones that I had been faced with. So the kinds of things they’re focusing on are a bit different. And so I thought it made sense to do these interviews to really get all these different perspectives. And you can also see a lot of commonalities. Those are the times when it’s like, oh, it’s not just Roy’s random opinion, but really these are the same types of things that everyone’s seeing. And then you see those differences. I also thought it would be, hey, this would be pretty easy to get cool content.

And of course, that’s also probably a poor estimation on my part because it does take a lot of effort to record those interviews and then transcribe those interviews and everything. I tried all the transcription tools out there also, and none of them were very satisfying.

[09:45] Tim: I self-published the Should You Get a PhD book. I guess it’s coming on 8 or 9 years ago now. I didn’t do audio interviews. I did email interviews. The difference between people promising to complete an interview and being willing to actually sit down and answer the questions and email them made it a much harder process than I anticipated. I really like that format of things. There’s a ton to be gleaned through talking to multiple people about the same rough topic and getting their different perspectives. And there was a book, maybe Data Scientists at Work or something, some years ago that was like just interviews with data scientists that I always found really helpful. And I think that’s a really solid addition to this book and I’m sure that’s gonna be helpful for people.

[10:36] Roy: Yeah. I’m considering sort of continuing a series there either with sort of podcasty stuff or some blog posts because I think there’s just so many interesting stories. I had at least one other person that was lined up that I didn’t get around to in time for the book, but we’ll see what happens there. Maybe I’ll add it in version 1.1 or whatever of the book.

[11:01] Tim: Is this the first book that you’re aware of on this topic specifically about data science? And if so, which I think the answer is probably yes, are there other blog posts or Medium articles or talks that people tend to cite on this?

[11:15] Roy: So there actually is another book out there that is about hiring data scientists, and it was written by DJ Patil in 2011. So, you know, that’s—

[11:28] Tim: Oh, right. I remember that.

[11:30] Roy: That might have been before they had books, but—

[11:33] Tim: That’s back when data science was the sexiest job of the 21st century.

[11:37] Roy: Yes. And it’s actually, I think, like a free ebook that’s like 18 pages long or something. So it’s a bit different in scope. And obviously a lot’s happened in the last 10 years, but there’s still a lot of good information in that book. There are a couple of sites out there that have some in-depth blog posts. They’re mostly from recruiting agency type people. But overall, I think there is just a dearth of material. I mean, if your goal is to get hired as a data scientist or machine learning engineer, there’s a ton of stuff online, a ton of resources. But if you’re on the hiring side of things, there’s just not that much. And that’s one of the reasons that I decided to write this specific book.

[12:19] Tim: Yeah, and like you, most of my experience has been in relatively small companies, startups or other smallish companies that don’t tend to have as robust of a hiring process as you can develop in the bigger companies when you have just a lot of resources and people devoted to it. I’ve actually not been involved in that much hiring in my career, but I’ve been surprised on several occasions where a day before an interview, I’m said, oh, go interview this person. And you try to find out, oh, what am I supposed to be accomplishing here? And well, is this somebody we should hire? No sort of framework or anything.

[13:03] Roy: I would call that kind of an anti-pattern, if you will. One of the things I stress in the book is that you need to have a process and you need to train your people to follow the process. And I also mentioned, of course, your process will not be perfect. You’ll need to iterate on that, but you still want to stick to what you’ve got. I think a lot of people who haven’t spent that much time in the hiring world or maybe just haven’t spent enough time thinking about it, they feel okay winging it. And sometimes maybe that’s okay, but for the most part, especially when you have these high-volume situations, especially if you have applicants who have potentially many other places they can go, that’s probably the wrong way to do it. You end up kind of wasting people’s time.

[13:54] Tim: Maybe I don’t want to admit that most of the jobs I myself have taken have been at companies that have anti-pattern hiring processes, but that’s probably been the case.

[14:06] Roy: I mean, it’s common. And especially in the startup world where when you start small, it doesn’t make sense to have an absurd amount of process, right? You can’t be a bureaucracy at that level and really get to where you need to be because you need to be so flexible so early on. But sometimes there are some basic kind of guardrails that you can put in, I’d say, and just to not do really dumb things, which can happen. It really depends. I also mentioned in the book things about the scale of the organization. And one is just as a hiring manager, at some scale, you usually don’t have nearly as much flexibility in how you want to run your process, right? Because there are established things. The sort of highest end is like, I talk about this in the part about decision-making, is that at some point, your company is big enough that they’ll say, we should have a hiring committee that is sort of higher up than any team hiring because they want to standardize things, right? That’s a very, very common thing to do. And at a certain size, that makes perfect sense, but at most smaller sizes, it probably doesn’t.

[15:21] Tim: So I guess I also don’t know anything about any broader literature in this space, but if there are books out there, and I assume there are, on hiring maybe just software engineers or other kind of technical roles more broadly, What is distinctive about this book being specifically for data science, machine learning engineering that is gonna be useful for people in those industries?

[15:46] Roy: Well, I think there’s probably 2 things. One of them I can sum up in a sentence, which is that at this moment in time, the volume of candidates for data science and machine learning roles seems to just be much, much higher per opening than your typical software engineering opening. So you just have to think about the process in a different way because you have a different scale that you need to contend with. The other part is just the specific roles this is about. It’s talking about the specific skills you need and kind of landing on what those roles are. And some in sort of the book, the stuff that I’ve written, and some also in the interviews, you talk about some of the differences between those 2 interviews, let’s say the software engineers versus the data people. And some of it is about the anti-pattern of applying software engineering interviews to the data people, which I’m sure that you’ve seen that. I’ve definitely seen that, especially in the earlier days.

[16:53] Tim: Some people will be familiar with a podcast that I was on that I’m very proud of, which is with our friend Joel Grus and Andrew Musselman on their Adversarial Learning Podcast. It’s been some years ago now. We did an interview about bad interview experiences we’ve had on the interviewing, on the candidate side of the table, not on the hiring side.

[17:14] Roy: Right.

[17:15] Tim: And we had all been through the wringer several times around, but I wondered if in writing this there are takeaways on ways in which a candidate can improve the process for themselves, you know, even going into a company that may not have all these things figured out and may not have a good process in place.

[17:41] Roy: Right.

[17:41] Tim: Are there questions that a candidate can ask or approaches a candidate can take to make you know, a better experience for them, and which ultimately is going to be a better experience for the company as well.

[17:49] Roy: Well, I think it’s— I think this is a difficult thing to do. You know, there’s so much variation in the companies and their processes that they use that it’s really hard to know. But probably the best thing is just for candidates to have a better sense of the challenges that those companies are facing. Because what each of you is kind of optimizing for is very different. And I guess I would say when, you know, having been on the other side of the table and looking at an opening where you’re getting hundreds or thousands of applicants potentially, you know, you realize that even a really, really good person won’t necessarily make it to the final stage or get an offer or whatever.

And a lot of times it is not the candidate, you know, it’s the competition they’re seeing, because the size and volume of that applicant pool raises the chance of also errors that people make, right? Whether it’s stupid mistakes like, you know, not emailing someone back in a timely manner or at all, or just some more fundamental flaw in the process. But I would say it’s important to understand that these are very, very noisy processes that you probably shouldn’t take personally. And that’s not to say that there aren’t bad companies, bad actors, but even the best ones are going to do stuff that feels unfair, at least. So in that sense, it’s maybe be a little stoic, I guess.

And then on the other side, the action, I guess, is to apply widely and to also try to put yourself in a position where the companies might take more notice, such as if there are companies that seem reasonable to you and you have people that can refer you to help you potentially get skip a step in the process or just gain slightly more attention from the hiring managers. It’s really hard. I mean, on the candidate side, I think one of the things that’s in my book that’s probably frustrated some candidates is that a process I’ve used and a process I recommend if you’re in a very high-volume, low-resource scenario is to not use the resumes as a screening filter. And the way that the sort of mechanics of that is basically, you might do like a very cursory look at the resume just to say, is this at all in the right ballpark?

And then if so, then basically for the next step, give them a relatively lightweight technical assessment. And then later on, you still use the resume, but not just like as a screener. It’s more as, hey, this is a person who seems to have the basic qualifications, and then maybe a phone screen has been done next. But now we have this resume as additional information. So it’s sort of a flipped-around way of how you use your resume. But on the other hand, a lot of candidates put a lot of time into their resumes and all that stuff, so they might feel unhappy when they hear sort of the advice of ignoring resumes.

[21:20] Tim: Yeah, I mean, my resume is generated on GitHub Actions and generates both an HTML and a LaTeX-built PDF file. So I really, I need you to respect it when you look at it.

[21:34] Roy: Yeah, it is so hard. I mean, it’s the same thing with profiles on GitHub or wherever, or cover letters. I did a poll recently on Twitter that was basically, in response to someone who asked the question, or they gave the advice, you should put a lot of thought into your cover letter, and that’s what gets people’s attention who are hiring. And the cover letters are something that I seldom look at. So I put a poll up just asking because I didn’t know the answer, which was, how often are you as a hiring manager looking at cover letters? And the answer seemed to be that it wasn’t very often, at least in the small sample I got in that poll. And that’s something that can be frustrating too, is just you’re putting a lot of effort in there and then it’s not really being seen. And honestly though, this comes down to volume, right? Tim Hopper is a very experienced senior-level person. So hopefully he is looking at these staff, staff-plus roles where there are very few peer applicants, candidates there. So, at that point, it should be that people are gonna love your LaTeX resume and all the tweaks you made.

[22:49] Tim: Did get a compliment from a recruiter on it one time, only in that it was very— it’s like very clean and simple. The PDF form is not very frilly, and she appreciated that.

[23:03] Roy: She just said, well, next time I’ll just send you the raw LaTeX file.

[23:05] Tim: So, can you go into a little bit of what you mean by a very lightweight technical screen? What is the details of what something like that might look like?

[23:17] Roy: Sure. I’ll give you a little bit of the background of how I got there because I think that’s kind of interesting, which is basically at the first startup where I was hiring, I decided to send everyone a technical skills assessment. And in that case, I sent people a dataset and then basically asked them to build a predictive model and then do a write-up. of, you know, what you did, what you found, why you did what you did. And, you know, overall that worked very well. We probably had for one position, we were looking at 100 to 200 applicants maybe, and a large number of them just simply don’t do the assessment. This was one where I said, you know, here’s the material, turn it back in in a week. And honestly, I didn’t even care if it was exactly a week or if they asked me for more time. I always said, fine, that’s no problem.

And that worked pretty well, I think. Now, when I went to my next company just a few years later, I decided to do essentially the same process. But what happened was the volume of applicants to these roles had just multiplied manyfold at that point. These are both startups that essentially no one’s ever heard of. So it’s not like we had some big brand name out there. It was just the size of the candidate pool in general had just increased so much. So at that point—

[24:42] Tim: And was that through job aggregator sites like Indeed or something like that?

[24:45] Roy: Yeah. I mean, we were both, for both of these roles, we’d apply at, or, you know, advertise on Indeed, LinkedIn, these sort of general sites. And, you know, immediately it was clear that if I sent this assessment out to all comers, that there was gonna be no way that I could actually handle the scoring and still sleep or whatever. So the first thing I did was I put in a resume screening process, exactly what I just said not to do a second ago. But in this case, what I did was I made a relatively simple scoring rubric for the resumes and just going through them, that was a lot of effort. And then doing the actual scoring, In this case, actually at that point outsourced the scoring of these assessments to another data scientist who had been consulting with us but was very good at what they did.

But even then, there was a lot of overhead in sort of coordinating things. There was overhead coordinating just sending the material to the candidate. We actually, instead of giving them a week at that point, I think we were giving them something weird like 6 hours. There was a disagreement between my boss and myself about how much time we should give them. And this is the kind of thing that you always run into is people have very strong opinions. My boss at that time felt strongly that only people who didn’t have a job would be able to do, or people who didn’t have a job would be able to spend an unlimited amount of time on a week-long assessment. And so we did some weird kind of compromise on that. But anyway, it was still a huge hassle.

And so the next time around when I did hiring, decided to switch to an assessment that was mostly automated. And this is where I call it sort of a lightweight assessment, even though I’m sure some people would just not consider it lightweight specifically. But basically at this point in the overall interview process, this acts as an initial technical filter. And for machine learning and data science type roles, the main things that I put in these assessments is looking at some basic data analysis, which also includes very relatively straightforward ability to read in some data files and kind of munge them and put them together. And then also some— and then basically asking them some questions that are relatively straightforward, starting with like, what’s the mean of this column?

And then getting up to something very complicated where they’re doing a lot of joins and group bys and things like that. Nothing crazy, but the things that most data scientists, analysts, and machine learning people do all the time. So it’s pretty straightforward. And then some relatively basic programming. All the roles I’ve worked in have all been Python. So that’s what I’ve been using there. And then finally, some things around machine learning. In my last role, actually hiring there, decided to basically have the candidates write an essay about machine learning, which sounds a little strange, but the hypothesis was that we could probably get a lot of the signal about what they knew about machine learning by just asking them, what would you do in this scenario, rather than having them try to build a model and things like that.

So in that sense, it’s a relatively lightweight We’re not having them go too in-depth. We want to see that they have some of the basic skills. This was an online test that was mostly automatically scored, and the applicants were given a time window of a couple weeks in which they could take that assessment at any time. And then they were given like 90 minutes to just complete the test once they started it. So, If you were in a scenario with just a huge volume of applicants, on the one hand, you want to select the best, most skilled candidates. But on the other hand, you also, you just simply need to narrow that funnel.

And if everyone who applies is, I don’t know, whoever you think is very skilled at technical stuff, let’s say Elon Musk, whatever, you might have 1,000 Elon Musks. in there, but you literally don’t have time to interview all 1,000 of them, right? So you have to narrow it somehow. And there’s jokes about people throwing resumes down the stairwell or whatever and just taking the ones on the top couple steps. And the reality is you need to get it down some way, but also you hope that that one is sort of correlated to some extent with skill at that point. And So that’s what I run at that point when you have these very high-volume scenarios. And it’s not in-depth.

I always recommend to do another round of screening on technical skills because I’ve definitely seen, even in the very more in-depth sort of home, take-home tasks and whatever, I’ve seen people who would get through that and then you reassess them and you realize that Something doesn’t add up. And the same thing with these kind of more automated screens is that sometimes you’ll get people who happen to be able to do well or potentially cheated. I’ve seen a lot of cheating. And then when you, when you get them to an onsite or whatever and ask them some other questions and it’s clear that they, they don’t have what it would take to actually do the job.

[30:22] Tim: When you do a process like that, did you have a goal of a size or percentage that you were trying to scale it down? Or maybe that’s just a kind of a slider that you can say, so take assessments up to this.

[30:35] Roy: So I do talk about specifically that in the book. You know, it is trying to estimate your sort of bottlenecks in your process. And, you know, what— at what point are you limited by the capacity that you can deal with? And, you know, the most difficult limit there is going to be when you’re doing in-person interviews or team interviews or whatever. So I do suggest that people come up with these rough percentage steps that they want to take at each level. And then, you know, I mean, one of the general things I just talk about is that you need to revisit your process and you need to make adjustments based on what you’re seeing. So when you’re doing the technical assessment, if it’s— and you have some sort of score on there, you’re going to set some threshold basically for pass, no pass.

And you need to keep looking at that to see if it makes sense and then get down to where you need to be. But also just more generally, you need to be aware of where the bottlenecks are. One anecdote I mentioned in the book is that at one company I was at, one of the co-founders, suggested that he talked to promising-looking candidates, or if we had very promising-looking candidates that applied or came in somehow, that he get a chance to talk to them just to really get them excited about the company. And candidates responded really well to this. They really liked it when the C-level co-founder gets on the phone with them and tells them, this is what we’re doing, this is why it’s exciting. And of course, that’s also was specifically a person that was very good at selling what we were doing.

And, but the problem was that this is a person who had an extremely, extremely full schedule at all times. So it became a bottleneck in our process, sometimes delaying our candidate by a week or two or maybe even more. And so eventually I just had to go in and have a discussion with him and say, look, you know, I think this is There’s some value to this, but it’s just hurting us too much in our timeline and everything. And so he agreed that we just not do that anymore. And so that’s the kind of thing that I hadn’t really thought much about it. I think when I agreed to that initially, I hadn’t yet gotten a taste for the scale of the applicant volume yet. So it seemed fine at the moment, but then we started seeing that and You just have to kind of take a hard look and do things differently as needed.

[33:13] Tim: Something you alluded to at the beginning of this conversation that’s kind of perennial hot topic is the issue of different titles. And we have this thing called data science that’s been around for about 10 years and the meaning of it has shifted and split and you can go to one company and it means one thing and another company it means something else. We now have machine learning engineer and MLOps engineer and AI engineer, all these different types of titles. To what degree does that matter in your hiring process, or how are you thinking through those different taxonomies these days in terms of the hiring process?

[33:59] Roy: Well, the broadest advice I give is that the most important thing you can do is to crisply describe what the role is you’re hiring for. And the reason I say that is because even if the title doesn’t fully convey what it is, and it probably won’t, then at least your description can hopefully kind of get you there. And memes aside about people not reading job descriptions, I think that’s pretty important. You know, on the other hand, I think I think that the differentiation that’s starting to come about is important for the field, for people to understand what the relative different roles are.

There’s still a lot of disagreement on what each title means, but I think that’s probably a better— we’re probably in a better scenario at the moment than we were 10 years ago, where the data scientists had 100 different potential skills or domains or whatever you want to call it that they might be potentially focusing on. And some people were doing all that stuff. Those were the crazy unicorns. But then other people, they might both have that title, data scientist, and when you actually ask them what they’re doing, the actual work was very, very different to the point where any given data scientist couldn’t necessarily do the the work of some other given data scientist, right? And I think we’re still there to some extent.

I mean, I talk in the book about if you have an early company and whatnot, you may, or if you’re doing things that are kind of along the lines of consulting, you may really want sort of generalists that are very general to do a lot of different stuff because a lot of different stuff comes up. But I think overall that these descriptions are We’re getting there towards better descriptions. I’m sure that things will keep evolving, so it’ll always be a little bit hard to pin down, but I think that’s been good. I think there’s, in the people I know, for whatever reason, most people I knew that were doing this 5, 8, 10 years ago, they had that title data scientist. And I’d say most of the ones that I know, for whatever reason, are mostly now more on the MLE side of things.

That may be completely a sort of sampling bias from just the people I happen to know, but that’s how it is. But I guess important there is that a lot of them, they would specifically be looking for a role called machine learning engineer nowadays rather than data scientist. And so I think there started to be enough kind of divergence there that people are starting to think of the data scientist title more closely aligned with analytics versus, say, the machine learning and sort of at-scale heavy volume production machine learning, at least. Not the case everywhere, but that’s what seems to be going on. So I’d say in that sense, I think it’s important to try to understand the role, look at what titles make sense, but also kind of understand a little bit about the market.

So, you know, if you keep getting a bunch of people who are mostly oriented towards experimentation and statistics, but you really need someone who’s a deep learning specialist, then you might have the wrong title, for example.

[37:21] Tim: Unfortunately, a lot of companies just do a surprisingly bad job of thinking through what they’re actually trying to hire for. I know I’ve been an interviewer in cases where my leadership above me has not really thought through that. But I’m pretty sure I’ve also been a candidate on more than one occasion at companies that haven’t really thought through that. So, I mean, yeah, it can be frustrating. Yeah. And just the basic facts of this is a costly process to try to hire people. You should really sit down and really try to hash out what you’re actually looking for and not just copy and paste somebody’s job description and think it makes sense.

[38:03] Roy: I mean, in one of the interviews, I believe it was with Julie Halleck, I could be wrong. I reread my entire book in the last few days, but it was kind of long. And I think that the advice was basically that you just need to spend a lot of time on this. You’re not going to get it right unless you really spend a lot of time and effort. And I totally agree with that. It’s easy to not do it well. It’s very, very easy to not do it well. And if you’re going to do it well, it’s just going to take a lot of blood, sweat, and tears, unfortunately, probably mostly your tears, but also other people’s tears.

[38:43] Tim: Before we wrap up, anything more you’d like to talk about the book? I’d love to talk through every bit of it, but I think we should leave something for people to go buy the book and read.

[38:55] Roy: One of the interesting things for me in this book was that it was self-published, like I said. I never thought about writing a book necessarily, but it’s interesting to see what was out there as far as self-publishing. I kind of went with the first option I saw, which was Leanpub, but so far, I’ve been pretty happy with that. But I followed a very programming-like process. I wrote this book in Markdown, pushed the changes to GitHub, and then Leanpub would go down and grab the changes and build a new PDF and the mobile formats. So that was kind of interesting. I tweeted some, maybe I’ll write a blog post about it, how I also did some writing on my phone, hooking up a keyboard and stuff when I was on the go. So that’s a little strange.

[39:47] Tim: But you didn’t dictate any of it into Google Glass, did you?

[39:52] Roy: I didn’t do that in the shower. But it has been interesting. And I’m interested to see the feedback coming from the readers and what kind of conversations this starts with people. So that’ll be good. I’m also just kind of coming off doing a bunch of consulting and then writing this book, I’m about to go back into the job market myself. So we’ll see how that goes. Yeah, why don’t you take the chance to expand on that a little bit? Sure.

[40:23] Tim: If you wanna tell people what you’re looking for.

[40:26] Roy: Sure. So I am really looking for the opportunities to build and lead machine learning teams or potentially data science in an organization. I guess it depends on the size of the company. Smaller companies, I’m probably looking at sort of VP+ level, and then maybe midsize, more director. And then at some of the larger tech companies, I’m happy to go kind of at a more mid-level manager just to get a lot of experience at different scales. I kind of, when I was thinking about these startup ideas, my other interests are around climate change. preventing climate change, I guess I should be specific, and around privacy. I’m known to be a privacy enthusiast. And so, you know, I’ve just been— I’m going to start looking, see what’s out there. I’ve talked to a couple of people already, but kind of— I’m kind of open to different industries.

But I’m hoping to find something that’s either remote or in Houston, Texas, which is where I’m located. And what I was going to mention before was that one of the tweets that I sent out recently was just about when I actually talked to people, and I’ve talked to a couple people in the past few months or a couple companies that I’m also looking at their hiring process because I know that’s going to be a lot of what my job is. So sort of, you know, taking notes about how they handle things and how smoothly it goes and how they inform the candidates of the process and all of these things, you know, those leave an impression on me that make me get a better feel for how difficult will it be to do my job if I actually go to work at this place. Right.

[42:24] Tim: Yeah, those can be varied experiences as we all know. Despite being a privacy enthusiast, if people want to get a hold of you, they can go to your website, roycoding.com, C-O-D-I-N-G dot com, and your email address is there and they can talk to you about job opportunities. And you’re Roy Coding on Twitter?

[42:46] Roy: Yes, and I currently, I’ve never done this before, have my DMs open. So if you want to slide in and ask me about hiring or you know, offer me a role I can’t refuse, you can potentially DM me, but also email. That’s probably easier for an old-fashioned person like myself.

[43:05] Tim: Excellent. And Roy’s book is available at dshiring.com, DS for data science hiring.com, which will take you over to Leanpub where you can get a copy of that. And it’s just ebook only at the moment, right?

[43:20] Roy: It’s only ebook.

[43:22] Tim: Okay, very good. We recommend it. I think people who are listening to this podcast, if you haven’t been involved in hiring, you probably will at some point, and I would suggest picking this up. And if you haven’t been involved yet, when you are involved, you may have the opportunity to shape how your company does it for the good and not be totally blindsided as I’ve been a couple of times in my career. I think if you came in with the knowledge from this book in hand, you would be able to offer your company a lot. So I think it’s going to be useful to a very wide audience, not just, you know, hiring managers, current hiring managers.

[43:59] Roy: I think there may even be value there for the candidates as well who want to just have a better understanding of how these processes work.

[44:08] Tim: Excellent. Roy, thank you for coming on Into the Hopper.

[44:11] Roy: Thank you for having me, Tim. Best of luck to you.