Last fall, I participated in a job-seeking fellowship called Insight Data Science. The official program lasts for seven very full weeks, but the job-seeking process continues afterwards for anywhere between a few weeks and a few months.
I was one of the lucky ones to get a job in the month following the program. So based on that fact alone, you might imagine that this will be a glowing review of the program’s success.
But while I’ll be focusing on my own experience, which as the title suggests was positive, I think it’s a bit more nuanced than that. Insight is much more than just a recruitment agency, and as I’ll talk about below, I think I’m better able to do my job having gone through it, as compared to simply getting a job out of my PhD.
Moreover, the community it builds has given me a fairly close-up look at the variations in experiences that people in the program go through. Many of my fellow Fellows have jobs, but many are also still looking, two and a half months (including holidays, to be fair) after the official portion of the program.
A few more clarifications: I was in the Fall 2018 Data Science (DS) program of Insight Data Science in Boston. Our program was run concurrently with the all-but-identical Health Data program, as well as the fairly disjoint but also co-located Data Engineering program.
With all of that said, here are the five things that Insight Data Science gave me.
1. A New Pace of Work
In the first four of the seven weeks of the program, we each formulated, developed, and executed individual data science projects that we could showcase for the companies that we want to work for.
Four weeks might not sound like long, but it was actually subdivided even further. We spent the entire first week churning through idea after idea, discarding all of them until we found one that was feasible, reasonably impressive, utilized relevant data science tools, and didn’t fall into categories that had already been well-trodden by previous sessions’ fellows.
On the other end, our entire fourth week was, at least in theory, devoted to polishing our 5-minute Google Slides presentation of our work. That left just two weeks to get the entire project done, start to finish.
But even that wouldn’t be a quick enough review cycle. The second week alone, we spent developing a minimum viable product (MVP) version of our work. We then spent the third week building off of our MVP to a fully-functional data product.
Of course, this was intentional; it wasn’t simply the product of organizational bloat. Academic work is often so long-term that employers rightly fear that any PhDs they hire will only be able to get something done if they have months of time to themselves. Very little gets done in industry by solitary geniuses like that.
The MVP step was also incredibly helpful in changing my own mindset. I’m rather perfectionist — if I notice problems in my work, I’m highly motivated to fix them as soon as possible, putting off further development. This can lead me to carefully polishing only a portion of the necessary work without seeing the whole pipeline through to completion. Maybe another component’s failure would doom the project, but I wouldn’t have time to work on it because I’d focused so much on perfecting other components.
I get the sense that perfectionism used to be more valued in industry, but times have changed as the pace of innovation has quickened. Tech companies are more and more adopting business practices that fall under the umbrella adjective of agile. Some people even refer to it as a noun: “Bringing Agile to the Whole Organization,” as if, you know, agility wasn’t already a word.
I was even able to see the wisdom of that approach in the midst of my project. Once my brightest idea — an app that would assess the viability of commuting via bikeshare and allow you to estimate when the bikes at your station would run out — unfortunately fell into the grey-list category of “things these companies have seen too many Fellows present on recently,” I took a hard look at the relatively small number of consulting projects that the program directors had solicited.
In that list, I was fortunate enough to find one from a fascinating startup called Copia which is trying to tackle the gigantic issue of food waste. They’ve identified that some of the biggest contributors to the problem are large businesses — caterers, universities, and so on — which end up with large amounts of perfectly edible leftover food after their events that they normally would throw in the trash or compost. Copia partners with those organizations to donate that food in less than half an hour to nonprofits in need, and they were looking to build out their predictive analytics platform.
That’s actually about all the information that I initially got from them, apart from the data itself. It was my responsibility, in consultation of course, to define what we would do together and then build that system into their existing capabilities.
To get it done, we met over Zoom about every other day for those 2-3 weeks. I had to take an MVP approach because I needed to have an idea of whether my approach would work before building it out fully. This let us quickly find snags, like that the Python package I wanted to use wasn’t supported on the online data visualization platform they would be displaying the data on. We then had to decide whether we’d work with an offline version or use a simpler model. Those sorts of decisions would affect my whole workflow, and we needed to sketch the whole pipeline to recognize the problem, rather than simply diving straight into coding.
In its ideal use case, the MVP approach encourages you to work on whatever is most lacking at the moment. You develop each component to its barebones minimum first, and then improve the weakest link each time. That way, if it just isn’t possible at all, you realize this quickly, and if certain components that you didn’t expect to be crucial turn out to be, you can devote more of your time there.
I’d also be remiss if I didn’t mention one of the biggest strengths of the program, the four technical advisers we shared as a group. Mine was Jerry Chen, an Insight alum a couple years prior, and he was able to offer me plenty of concrete feedback and suggestions as I developed my project. Throughout the course of the program, various alumni would visit and offer their fresh eyes to our projects and presentations, which was also incredibly helpful, but Jerry was always my go-to-guru.
2. A Snapshot of the Industry
In the middle of those first four weeks, we were, ahem, constantly interrupted by one-hour presentations given by data science teams at the couple dozen or more companies that were looking to hire data scientists in the Boston area. Of course, we had to balance the need to build an impressive demo with the need to think about where we’d actually like to work, and this was also an intentional attempt to mimic the atmosphere of actual data science work.
These presentations were one of the aspects of the program I was most looking forward to. When I had done my half-hearted job search before applying to Insight, I hadn’t come up with a very long list of companies. There were the tech giants everyone’s heard of, and a few smaller companies that people had suggested to me.
By contrast, the companies looking to hire at Insight tended to fall into four main categories:
- Tech companies, led by the rapidly expanding furniture retailer Wayfair (that is, a furniture retailer which is rapidly expanding, not a retailer of rapidly expanding furniture).
- Consulting companies, or their data science arms and acquisitions.
- Startups taking on a wide variety of exciting but risky endeavors from agriculture to marketing.
- Health-related companies with a wide range of therapeutic focuses, Boston being the biotech hub that it is.
In addition to this bird’s eye overview, we also got a rare in-depth glimpse into the internals of these companies and the specific roles they had in mind. The hour we got together could range from about 10% presentation and 90% Q&A to 75% presentation and 25% Q&A. Companies that had hired in the past would often bring back some of their Insight alumni, who we could trust to describe the company from an approximately similar perspective.
To be honest, at least for the companies at our program, I was somewhat disappointed that there weren’t more companies with clear social purposes like Copia, the startup based in SF that I did my project with. I think this is sadly primarily reflective of the economy as a whole, though — if you’re not making some rich person richer, you’re not likely to find a well-paying job in it.
3. A boost in confidence
The last three weeks of the program, spilling over into the interview period following the official end, were primarily composed of two activities: Going to each of the companies and demoing our work for them, and preparing ourselves for the subsequent interviews.
The interview prep portion was fairly extensive but also quite unsupervised. Many of the questions didn’t have answers, which made it hard for us to tell how we were supposed to be answering.
If you know me at all, you can see how I’d be frustrated with this. Some of my fellow Fellows decided to compile our own answers, or at least tips on how to answer, and I contributed answers to the probability questions. We wanted to at least be able to build off of the collective knowledge we did have, which seemed to be in line with the overall purpose of this time.
In my opinion, one of the best structures for interview prep they provided us with was an “Interview Simulator” which we participated in every Friday morning. We were divided into groups of 6-8, and took turns asking single questions that we’d prepared on a variety of topics. Preparing that question gave us a target topic to study for that week, and answering all of them gave us exposure to our weaknesses in different areas.
Other than that, those few weeks were pretty unstructured. One reason is that each of us had different areas we needed to work on, so we naturally went different directions. For instance, I found that I was somewhat weaker on coding in base Python, having used Mathematica and R in my PhD work, so I spent a good chunk of time just going through challenges on HackerRank.
In general, this period of time also boosted my confidence heading into interviews. Prior to Insight, I think I had seen my experience and knowledge as too theoretical for a real job, and as I worked through the materials they had prepared for us, I became more confident in my abilities.
In fact, I probably got a little too confident. I went into my first on-site interview with high hopes of clearly demonstrating to my interviewers that I had the technical skills and relevant experiences that would make me an awesome member of their data science team. In the process of showing off my awesomeness, though, I rubbed some of the team members the wrong way, as one of their team members very graciously explained to me when he told me that they regrettably could not extend me an offer.
4. A network of compatriots
That was definitely a tough phone call to take. Fortunately, it was during the day, and my fellow Fellows were there to comfort me after they asked how it went. “Their loss.” If I had been on my own, I think I would have taken it a lot more poorly, and probably lose motivation to keep up my job search for a while.
But I knew I wasn’t alone. Plenty of other fellows had been excited about particular companies only to be rejected at various stages, often without seeing any particular reason. The process is quite chaotic, and so much of the variability depends on the timing of when you look.
Through the chaos, though, we’ve been there to support one another. We would applaud whenever a group would return from a demo, even if they tried to sneak back in without anyone noticing. As we started to scatter after the program ended, we used a Google form to post progress steps in our interviews to our group message board. As people started to get jobs, we’d react with emojis to congratulate them.
In addition to the general encouragement, we also had the opportunity to teach each other throughout the interview preparation period. Of course, this came up naturally in informal conversations (“wait, explain this to me”) and in debriefing the interview simulator. But there was also a new structure in our session, Fellow-Led Workshops. Each of us prepared a 30-minute presentation (whiteboard or slides) to the whole group. Some people picked a general topic that they felt confident in, while others picked specific tools that they used in either their Insight project or their PhD.
When I was first considering whether to do Insight, one of the most common things that I heard was how good the networking was. Networking always felt like an icky thing to me, like sidling up to someone at a party and handing them your business card, or e-mailing someone out of the blue who happened to go to your alma mater 10 years before you did.
I still find networking done like that repulsive. But Insight, at least between fellows who are there for the same session, is networking done right. The bonds formed over shared experience, especially the emotional roller coaster that a job search can be, are a lot stronger bonds than the sort of weak connections that stereotyped “networking” usually forms.
5. A job!
At the end of the fourth week, we ranked all of the companies we’d seen so far. We were then assigned (by an algorithm, of course) 5-7 of them to go for demos. Almost every company got 4-8 candidates this way, with the big outlier being Wayfair, which got 17, coming in two batches.
I almost got my top five choices — Wayfair was my #4, and, well, 17 of my fellow Fellows ranked them higher. Instead, I got 5 of my top 6, which I was pretty pleased with. In general, I tended to favor the tech companies, both startups and more mature companies, and ended up with a good mix.
We’d spent those first four weeks doing the work and preparing presentations on what we did, including a slide at the end about ourselves, where we’d then say why we wanted to work at that company. We’d rehearsed our standard presentation countless times, but we had to come up with a different thing to say for each company, and of course in the process think about why we wanted to work there.
I never could sleep well the nights before those demos. To make matters worse, at the end of my first demo presentation, I completely forgot to say why I wanted to work for that company! Unsurprisingly, they didn’t invite me back.
The night before my interview with Kebotix, where I would ultimately get a job, I was pacing around our apartment. I’d given my standard presentation so many times that I had basically memorized everything I would say without having to look at any slides. I was timing myself, though, to see if what I wanted to say on my last slide would fit in the recommended 5 minutes overall.
I knew why I wanted to work there: It was a fascinating opportunity to combine my interests in math and chemistry that I didn’t know existed. I still wasn’t sure I wanted to work at such an early-stage startup; I’d always envisioned my journey to the working world taking me to a more typical corporate job. I might not have even applied on my own. But here I was, demoing with them the next afternoon, and I needed to explain why.
As I paced back and forth after midnight, I hit upon the way I wanted to formulate it. It went something like this: “The reason I’m interested in working in Kebotix is something that you’ll find on my resume, but which doesn’t feature too prominently. While at Caltech, I actually got a double-major in math and chemistry, and I did research in my summers in both. The reason I chose to go to grad school in math over chemistry was because I found that while math research proceeds at the speed of thought, chemistry research is as slow as the pace of your experiments. I love what you’re doing to try to get chemistry research to proceed at the speed of thought, and I’d like to help make it a reality.”
As I hoped, this made a big impression. Of course, they also liked the fact that I had a chemistry background at all, and in my first two months at the company, I can see how just speaking that language has been incredibly useful. When you work at a startup that size, everyone needs to be interdisciplinary.
After our demos, we started to hear our callbacks. I ended up hearing something from four out of five, but they all ended up fizzling out in different places apart from Kebotix. I was surprised at how dissimilar the processes were at different companies; there were several common stages like “phone screens” and “on-sites” but even those categories hid immense variance in terms of both content and length.
And now that I’m on the other side of the table, I can see why. We’re just now designing our interview processes for typical (non-Insight) applicants, and we frankly have no idea what we’re doing. In that way, it was even more helpful to go through Insight, not only to go through my own interviewing experience, but to hear from so many others about theirs.
The particular set of companies I demo’d with tended to be on the faster side of things, so my first week after Insight ended filled up fast. I had that first unsuccessful on-site that Friday, and then had my Kebotix interview the following Monday. They didn’t tell me much about what to expect except that it would both be a one-on-one interview and that I should prepare a 20-minute presentation on my PhD work. Hmmm…
So when it came to the end of the interview, having spent about an hour and a half going through half of my thesis defense slides amid their constant yet engaging questions, I was somewhat surprised when they told me that this was the end of the process; the next step would potentially be an offer.
That offer would come eight days later. By that point, five other fellows already had offers, including another who got hers simultaneously at Kebotix, my now-coworker Tanja. While Wayfair would go on to hire four Fellows from my cohort so far, Kebotix was the first to two. 🙂 We both got the requisite ~20 robot emoji reactions.
With an offer in hand, I encouraged the last remaining company to get back to me and move forward as soon as possible. Kebotix was inviting me to start the following week, even though it was the week of Thanksgiving! I also did my own research and tried to understand as much as I could about the startup world and the equity in the offer, so I could properly negotiate.
I’m not really sure how much I should write publicly about negotiations with my employer, but I’ll just say that I’m glad I took that time. I ended up needing the weekend and eventually starting a day after Tanja, the Tuesday of the week of Thanksgiving, a hardly noticeable delay except that it meant that she was the one to start drafting our onboarding procedures for future employees.
And that’s where my journey with Insight ended and my journey with Kebotix began. That Monday before starting, I came in to say my goodbyes, turn in my key, and share a last bowlful of chocolate-covered pretzels with the friends I’d made there.
Of course, it isn’t over for everyone. I don’t know if I’m 100% up-to-date on everyone’s progress despite the Google form and emoji system, but hiring naturally died down around the holidays and hasn’t really picked up since then. Not everyone is happy about how that’s gone, with second rounds of interviews and going back to applying to stuff normally. But most of us, job or not, are keeping in touch with each other; we’ve migrated most of our cohort over to WhatsApp and grab dinner every once in a while, and probably will continue that in the near future. As they like to say, once a Fellow, always a Fellow.
I imagine a couple types of people will find themselves reading this post. If you’re a personal friend of mine, I hope this has been an insightful look at a pivotal couple months of my life. If you’re also considering a transition from a PhD into data science, I would highly recommend checking out Insight, and I’d be happy to formally refer you or answer any questions you might have.
If you’re instead Googling for information on Insight and stumbled across this post that way, I hope that I’ve been able to offer you a bit more of an unfiltered and personal (if rambling) look at the Data Science program, at least as it existed in the fall of 2018 in Boston from my perspective. As with any startup, Insight is also constantly adapting, saving and expanding the good parts while discarding or modifying the bad, so take the details of what I’ve said with a grain of salt, especially the more distant the program you’d be considering is from mine.
You can find out more about the various Insight programs and apply at Insight’s website, http://insightdatascience.com.