My Journey from College Graduate to Data Scientist
The COVID-19 pandemic and the resulting economic fallout has generated devastating financial impacts to everyday Americans. This pain is felt especially by younger Americans graduating from college and those less established in their careers. Without a developed social safety net or much in the way of savings the COVID-19 recession is a setback many can’t afford. Through no fault of their own, even those lucky enough to be newly employed are likely to be paid less not only today but years in the future. As bleak as the present looks there are bright spots in the economy.
What is clear a few months into the crisis is that the COVID-19 recession is accelerating preexisting trends toward more agile technology companies and away from older established businesses. The dramatic stock market rebound since the March 2020 nadir is almost entirely due to the dramatic outperformance of technology companies. Today Apple, Amazon, Facebook, and Google make up almost 22% of the market capitalization of the Sp500 compared to 17.5% in February 2020. The dramatic rise of companies like Zoom, DocuSign, TeleDoc, Shopify and others is an indicator that the most promising present and future career opportunities are in disruptive technology companies. However, the gap between moving from college (particularly a liberal arts degree) to working in a technical role at one of these companies seems like a massive chasm. I undertook a similar journey and I wanted to share these experiences for those looking to make the same leap.
A technical career track was never my intention out of college. I graduated from James Madison University with a BA in Political Science. I spent most of my university years going to punk shows in basements, making paintings on cardboard because I couldn’t afford real canvas, and occasionally writing papers about topics like Jewish Pogroms in Poland or the influence of Reinhold Niebuhr. Today I’m 3 years into my position as a Data Scientist at Amazon writing Python code to build Machine Learning algorithms. I never went to a code bootcamp and I never went to grad school.
I graduated in 2012 from JMU and as I had expected, a 3.4 GPA and a BA didn’t open a ton of doors. Although the economy had recovered from the worst of the 2008 recession, the experts were predicting an 8% unemployment rate as the new normal. I applied to 200+ jobs and received two follow up interviews and one job offer. A side note— I also had an interview for a door-to-door sales job in addition to the two aforementioned callbacks. To put my situation in 2012 into perspective, I considered accepting the salesman job at first. To say the least, my first day shadowing another salesman at the company was a bleak experience, During my first (and last shadow) I was mocked, then shouted at by multiple residents in a Northern Virginia cul-de-sac before being followed, photographed, and then escorted out of the neighborhood by a rent-a-cop.
As disastrous as that experience was, I was well acquainted with lesser desired roles in the American economy. In between my college semesters I worked as a cashier at Target and a Korean grocery store. I knew about working poverty in the US but it wasn’t until that experience that I became acquaintances and friends with people in that position. I saw how hard they worked (often double shifts or multiple jobs) and witnessed their difficulty gaining an economic foothold. I was never that motivated in college but this was enough to push me to get a corporate job.
After the door to door sales fiasco, I interviewed and was hired at Clarabridge. Interestingly, the attributes that got me hired at Clarabridge had little to do with what I had done in college. Between my junior and senior years at JMU (2011) I formed a consulting company with some high school friends. We had almost no idea what or who we were going to consult for. After cold-calling and walking the streets of downtown Leesburg soliciting business we secured a couple of small business clients for which we built some basic websites using CMS. In addition to passing the Clarabridge analytical assessment test (for which I only needed to know some rudimentary Excel) my small business success is what won over my hiring manager at Clarabridge and ultimately got me my first job.
Aside from mastering the Clarabridge software, my role as a Business Consultant at wasn’t very technical. Their software processed text and assigned sentiment scoring, which we then layered by building “categories,” which fed into reports that showed how well a given customer was performing (for example, determining how online checkout is trending more positively over a period of 6 months). As a Business Consultant, I showed customers how to use the software and extract out findings, sometimes creating PowerPoint presentations to show executives. The first few months were challenging, but after 6-8 months the job felt quite routine.
During the following year (2013), I experimented with more complex ways to present our software’s insights and automate the most boring parts of my job. I got really interested in Excel and built a tool to generate my reports. During this time I was first introduced to the concept of “data science” which felt like an abstract and esoteric new field.
Back in 2013, it seemed like it would be impossible to become a Data Scientist without getting a PhD or an MS in that field. At that time, I didn’t know how to program and had little knowledge Machine Learning. I thought the learning curve was going to be too severe to overcome on my own. I had done some basic HTML and CSS in the past, but the extent of my programming lexicon went about as deep as my high school French (je ne sais pais). What I did understand was the potential to apply Machine Learning and programming to my day-to-day job to exponentially increase what I was achieving in my role.
By 2014, I had gained an understanding of the toolsets and requirements of a real Data Scientist. I was looking to make a career change but I didn’t have enough on my resume to get much traction with recruiters. I was hopeful to go into Clarabridge’s product management department, but unfortunately that role was quickly filled by another internal team member.
Around that same time, my company opened a position for a “Data Scientist” in the marketing department. In the early 2010s, Data Science was considered one of the “sexiest jobs in the 21st century.” The new CMO of Clarabrige made it her mission to hire a Data Scientist to help create marketing content. I decided to apply for the role because I was anxious to expand my role at the company. I reached out to the CMO directly and showcased some of the more novel things I had built using the Clarabridge software, including some infographics (how 2010’s is that?). The CMO was impressed with my creativity and gave me the job immediately.
I quickly learned that this role didn’t have much in the way of hard requirements. My entire mandate was to create data driven content. I had complete flexibility to pitch topics, collect data, and create content. This open-ended position gave me the freedom to explore new technologies and increase our marketing department’s potential. Since I was the only technical member in a marketing department I was given very sparse technical requirements on how to accomplish business goals. As long as I met deadlines to deliver blogs, scorecards, or sales funnel tasks, no one asked how I got there. I took advantage of this flexibility to learn how to program.
A lot of my projects were centered around creating blogs and performing sentiment analysis using the Clarabridge tool. Although the software had some built-in APIs to connect to Twitter, Facebook, or flat CSV files, I decided to curate my own datasets whenever possible. In practice, this meant building scrapers in Python to crawl websites. During this time I went throughLearn Python the Hard Way and took Andrew Ng’s Intro to Machine Learning. In my spare time I created some ML side projects (e.g. Predicting Best New Music on Pitchfork) and started reading academic papers on Arxiv. This was when deep learning was just beginning to take off for commercial purposes, word2vec was transforming NLP, Keras was still a standalone project and Theano was still used.
I worked in this marketing data science role for about two years. By 2016, I had gained some practical hands on experience with Machine Learning and Python. Some of the research and modelling work I created were featured prominently in publications e.g. (Oscar-winners tend to inspire one particularly powerful emotion, especially in men). I created the first applied machine learning model using Clarabridge sentiment within the company. Although I made significant progress, I realized that within this role I wasn’t able to bring Machine Learning models to scale in real production systems. I left Clarabridge that summer and my job prospects increased significantly. Although I wasn’t getting much attention from the top companies I had no issues obtaining interviews.
My original intention was to move to Capital One, but I didn’t make it in the final hiring round. Part of the challenge was that at this point the more established tech companies conducted technical interviews (whiteboarding, coding tests etc.), which I had zero experience with.
After final round interviews with three separate companies, I ended up working at a mobile app startup in Arlington called Mobile Posse. This was another fortuitous move. Although I didn’t get the resume cachet that a company like Capital One provided, I gained hands-on experience with big-data, cloud infrastructure, and production ML workloads. The start-up nature of the company meant process was second to getting workloads to production. All hands were always on deck. Even though I was a Data Scientist, my tasks blurred into data and software engineer disciplines. By the end of my time at MP, the A/B testing, modelling, and architecture work I helped deploy were generating a significant business impact.
In the summer of 2017, the tech world finally welcomed me with open arms. I was being actively recruited by top companies (Amazon, Facebook, LinkedIn, Palantir, late stage startups etc.) and I had to actively turn down interviews. After my previous round of interviewing I knew what to expect. I spent most of my commutes (1.5 hrs a day from DC into VA) reading Data Science from Scratch, deepening my knowledge of probability theory, and mastering the fundamentals of software development in my free time. Surprisingly, the interviewing experience this time around was one of the most rewarding parts of my career.
Although I never enjoy white board coding, compared to the previous year I saw how I had progressed. Coming from a non-math and software background, there is a certain feeling of belonging when established industry data scientists (often with STEM PhD’s) approve of your technical standards. This was an important inflection point for me. Prior to these interviews I didn’t really know if I could make the jump to a “real company.” After multiple offers with different companies, I accepted Amazon’s offer to work as a Data Scientist in August of 2017.
From 2017 to today, my tenure at Amazon has forced me to learn even more than I did at my previous four jobs. Compared to my previous roles, I feel a bit spoiled in that executing ML algorithm development is the core function of the job instead of the 10-20% of my time that it occupied in my previous roles. Over the last 3 years, I’ve built models for AWS customers across multiple verticals (example project of mine), in multiple Machine Learning domains and even published a public blog (Increasing performance and reducing the cost of MXNet inference using Amazon SageMaker Neo and Amazon Elastic Inference). Overall one of the rewarding aspects of this experience was being able to look back and realized that I navigated my own path to get here.
I’ll finish by providing some key advice for anyone looking to change their careers or expand on their skills in the job market.
Credentials get you in the door, outcomes get you hired. Getting a data scientist title was very important to getting hired again as a data scientist (duh). However, credentialism only gets you so far. Knowing how to use the right tools and technology to achieve discrete business outcomes is the differentiating factor.
Learn how to learn. Self-acquired knowledge and possessing the capacity to learn are essential to transitioning careers. Part of what enabled me to pursue this path was that I’ve taught myself things outside of formal settings. If you have diverse interests and continually learn it’s less taxing to branch out to new things.
Find your motivation. Motivating factors are different for you than me. However, you always possess agency to find what motivates you in a particular situation. At some points of my career it was financial incentive, in others it was more about proving myself to other people. I think it helps to have a competitive personality in this regard.
Take risks and push yourself to fail. Although I had doubts about myself on many occasions, it’s better to have someone else decide you aren’t ready. When I first started coding it was an absolutely horrible experience but I pushed through that barrier. You have to accept failure as a prerequisite to learning.
You have to enjoy it. Almost everyone I know at my job has some level of passion and appreciation for what we do. There is an almost aesthetic quality to certain code or algorithms that makes me happy.
Great write-up!