In our many conversations with people who have switched careers, a good number of them recall the anxiety and challenges of moving into a new field after funneling years of study into a particular subject. Alex Mentch, for example, began his professional track as a missile defense researcher and a PhD candidate in electrical engineering. Just recently, he spoke to his alma mater bootcamp Zipfian Academy about his decision to leave his narrowly focused graduate program for greener pastures in data science.
Tell me about your background. What kind of work did you do at MIT and NASA?
My background is in electrical engineering, focused on controls and robotics. I have a BS and an MS from Washington University in St Louis. My concentration focused on linear algebra, statistics, and stochastic processes - essentially applied math. Controls engineering is about making systems to regulate themselves and respond in ways you want them to, like autopilot or cruise control. I interned at NASA, University of Idaho, and MIT Lincoln Laboratory, and I really liked it. I also worked full-time on missile defense research at Lincoln Lab before entering a PhD program in electrical engineering at the University of Maryland.
How did you get interested in data science?
I liked what I was doing, but the career started to seem too niche. I wanted to work in broader fields that had applications outside of the narrow industry that I’d found myself in. Right around the time I dropped out of the PhD program, I went to a DataKind weekend hackathon where I worked on a project trying to find a correlation between nighttime light intensity maps of Bangladesh and local estimates of poverty. I realized that a lot of what I had enjoyed about my work was actually data science.
What did you do after deciding to pursue a data science career?
Data science was based on a lot of math that I already knew, but I needed to learn new approaches and tools. I spent the first summer after I dropped out of my PhD program doing Coursera courses in data science and programming. However, I wasn’t making progress as quickly as I wanted to, and didn’t feel like MOOCs would make me qualified for the field. One of my friends completed the Hackbright program so I knew programming bootcamps existed. I typed “data science bootcamp” into Google, and that’s how I found Zipfian Academy.
Why Zipfian Academy? Why not a master’s degree in Data Science?
Sure, I looked at the programs at Berkeley, NYU, and Columbia. With one year of PhD experience, I knew I already had most of the skills, and all the math that I needed. What I did need was to learn the right methods and tools, so I didn’t think I needed another year or two of school. A lot of the data science master’s programs seemed to be designed around the idea that you need X number of classes to get a masters, which seemed inefficient. I ruled out a master’s degree right away.
The difference between Zipfian Academy and MOOCs is that learning alongside other humans is really helpful, especially learning from other people who are experienced in the field. I tried to do Coursera courses like they were real courses in a college, but it didn’t work. I was also interested in the connections to industry at Zipfian Academy, which I thought would get me on the right track.
What was it like being in the program at Zipfian Academy?
"Zipfian Academy provided that sort of learning that doesn’t usually happen until you’re working in a job."
It was intense. We were there from 8 or 9am until well-past dinner on most nights - to the chagrin of the neighbors on a few occasions. I appreciated that the program was very focused and hands-on. The lectures were designed to get us started working on our own - they weren’t any longer than they needed to be. Because the program is focused on hands-on work, you make a lot of mistakes in the beginning, but you figure out how to solve them. It’s a really effective way of learning the material. You also develop the intuition that you need in this job - meaning a familiarity with the algorithms and tools that are available to you, and what kind of questions you can ask of the data. That kind of experience helps you do your job faster.
In a typical college or grad school class, you only apply the thing you learned in lecture to the problem given afterwards. But with a guided homework assignment like that, you don’t learn as much about how to discriminate between a set of possible approaches. Zipfian Academy provided that sort of learning that doesn’t usually happen until you’re working in a job.
The program was also very collaborative. We did 3 weeks of pair programming in the course. Even after that, we were still asking each other questions all the time and comparing approaches to solve problems. This helped us learn quite a bit from people who had different backgrounds, and therefore saw the same problems differently. The capstone projects we built were entirely independent, but I still ran things by people in the program with whom I’d worked most closely.
Tell me more about your capstone project. Why build a search engine for state legislation?
Originally, I was thinking about a project related to ALEC, an organization that provides model legislation to state legislators. The organization has members in most states but doesn’t publish its member list or the bills it writes. As a result of ALEC’s activities, state legislatures will have similar bills that are brought forward for discussion at the same time, but the true motivation for the bill isn’t obvious to the citizens of the states it’s being discussed in. I wanted to see if I could tease out networks of state legislators that often sponsor these similar bills. What I found was that getting state legislation is very difficult. I ended up deciding to build the tool I needed to eventually do the analysis I wanted to do. I was able to get quite a bit of metadata from the Sunlight Foundation, but mostly I built the project by scraping bills from state websites and applying natural language processing techniques to make them searchable.
What was the data science job search like for you?
My interview process wasn’t too bad. I mostly used connections from Zipfian Academy’s Hiring Day, as well as AngelList, and Linkedin. As often as possible, I tried to find a friend of a friend that worked at a company to get in, and sometimes I was contacted through Linkedin.
I was talking to both startups and more established companies at first. At a startup, I thought I’d learn more about starting a business than about data science. If I were the only Data Scientist, I was worried I’d be reinventing the wheel over and over. That was why I decided to focus on medium and larger companies.
The typical application process was some sort of initial phone screen, a take-home assignment, then on-site interview with algorithms, SQL questions, and product questions. I’m sure there are other ways they’re probably evaluating you, but you’re never really sure. The on-sites were pretty intense with about three hours of back-to-back interviews.
I didn’t have to prepare for interviews too much. I was planning to answer algorithm questions in Python, so I made sure I was comfortable whiteboarding Python. I also made sure I was ready to talk basic stats, could whiteboard SQL, and that I was at least a little familiar with the company.
I interviewed at Facebook, Airbnb, Tesla, and Uber, and accepted an offer from Facebook on their Product Analytics team.
How much did your strong math background matter in your transition to data science?
"I really don’t think there is one mold that you have to fit...Having a certain level of statistics and programming in your background is all that is necessary."
Math definitely helps and makes it easier. The graduate-level math I have is more of a luxury. I’ve met a lot of people from similar technical backgrounds to myself, but also a lot of people with backgrounds very different from mine. I really don’t think there is one mold that you have to fit. Every company has a different definition of data science and different projects for data scientists to do. It is definitely possible for someone with product and business sense - or someone coming out of business school - to do this job well.
Having a certain level of statistics and programming in your background is all that is necessary. From there, it depends on how you want to shape your career direction.
At Facebook, there are people coming from finance, science, engineering, social science and business backgrounds - it’s pretty diverse. Of the people that graduated college around the same time I did, some went directly into data science or analytics roles, others into finance or consulting, and others into graduate programs in a lot of different fields. The common thread is that everyone was more interested in data and technology than they were in their previous roles.
Do you think your experience is repeatable?
I was fortunate to have the right people at companies consider my application - it’s definitely harder if you don’t have connections. I think that’s another thing Zipfian Academy did for me. I know I got a couple lucky breaks, but I was well prepared and was ready when those opportunities came along.
From what I’ve seen, it’s hard to hire a data scientist, but it’s also hard to get hired. I’ve talked to a lot of people who are interested in data science but aren’t really sure if they want to move forward. I keep recommending looking into Zipfian Academy or something similar. It’s important to learn the things you need to get into the field as well as gain the connections necessary to get started. Companies often list a PhD as a requirement for data science roles, but the work doesn’t actually require it.
If you’re asking me if I were to do it all over again, would make the same choices? In that case the answer is yes. This was definitely the right path for me.