How I got here.

I arrived at Lawrenceville at 17, barely speaking English. It’s always kind of hard to explain what it was like.

I went to classes with such smart people and listened to unbelievably good teachers. Our classes had us sitting around a round table, debating with one another and learning this way. I also got to play hockey at a competitive level without having to put school on pause. You get to eat breakfast, lunch, and dinner with your friends.

I also learned how to speak English there.

It was pretty hard, though. We had school 6 days a week and the amount of coursework was insane.

I arrived at Colby wanting to study something math related. And I did.

I had the chance to take the two highest level stats classes my freshman year. My sophomore year, I had an independent study with a professor on topology where I learned way too much about “knots” and their properties.

I learned about Machine Learning, Probability theory, Linear Algebra, Bayesian Statistics to name a few.

But the most valuables things that I got out of Colby were outside of that.

The random classes I took like Public Speaking. The language requirement slowly turning into a semester abroad in Spain, where studying Spanish was an excuse to do cool stuff like take a wine tasting class. I never in a million years could have guessed I would graduate with a Spanish major along with my stats one.

But also, the friends I met. The crazy nights out with insane stories. The weekend meet ups across Europe to visit them and explore the world.

I had time to have fun, try things, and really get to know myself. Go Mules!

During my sophomore summer, I worked as a Summer Project Manager at NECG, which involved the inevitable crash course in doing whatever the work required. I worked on building a statistical model for an F100 company’s vitamins division to predict discount depth and sales volume. In the second half of the summer, I built a web scraper for an F100 healthcare client to assemble datasets of prospective doctors that had the required skills and equipment to sell the healthcare client’s products.

A few months after the internship was over, I ended up leading a working session with a data scientist and marketing manager from the same healthcare client, walking their team through what I’d built and how it actually worked so they could adapt it.

The part I’m proudest of came three years later. NECG had sold a follow-up project to the same client, and they wanted a new scraping system like the one I had built. My old boss at NECG ended up calling me for help because they struggled to piece it together. I ended up writing the blueprints for a new version of this system and sending it over.

¡Viva España! I spent a semester studying in Spanish, in Andalucía. An incredible experience that Colby College funded as part of their financial aid program. I traveled all over Europe, went to La Liga games, and met super cool people. My favorite memories are just wandering around the city after class and discovering the city this way.

I jumped into Prospector having basically never used Python, knowing only the basics of R, and somehow came out having done a full software-development-meets-data-science internship in it. And working at a small startup was cool because I got to work on the core systems the company was building.

Almost all my time went to one problem, which was to systematically pull data out of mining companies’ PDF reports, specifically the tables inside them. I was mostly on my own to figure it out, and I learned an enormous amount about NLP, Python, and data extraction methods that pre-date LLMs.

One of my favorite breakthroughs took weeks to crack. Many of those PDFs were generated from Word documents, and the conversion engines are riddled with bugs, so tables would come out missing or mangled depending on the tools used to visualize. The realization that finally fixed it was that you can’t reliably manipulate the PDF’s PostScript layer directly. Instead, you convert it to bytes and run your search at the byte level, and the broken tables snap into place.

The processes I built and improved cut more than 30% of the time off manual data processing work that was done by hand by full-time employees.

Mars & Co was my first job post college. I worked there for 2 years as a strategy consultant. Mars’ approach was generally very data-centric, and so my work was at the core transforming data into insights for our clients. I worked exclusively on projects for F100 clients, varying from doing P&L modeling of a disruptive competitor product & market trends, general strategy analyses, building models to predict all sorts of scenarios.

But the project I am the proudest of was a project with one of the biggest companies in the US where we were tasked with reviewing the entirety of one of that company’s supply chains that carried billions of $ of inventory. We initially used decently complex Excel modeling, but I pitched to the partner that I had seen sophisticated statistical methods in college to fix this kind of problem. Our goal was to evaluate optimal distribution center positioning and estimate delivery costs of “milk-run” deliveries. I ended up being given a few days to pull together PhD theses and just industry cutting edge papers to pitch something to them that I could do. They were satisfied with what I proposed, and I then was given a week to put it all together.

I ended up coding in Python simulations using genetic algorithms, different clustering techniques based on dimension reduction methods that allowed me to model traffic in a “Euclidean” distance system, and reverse engineer Uber transit costs for cost data to eventually be able to mathematically answer the question “What is the most optimal number of distribution centers and their placements across the USA to ensure X level of service and under Y delivery constraints”. My simulations ended up the backbone of this project that had billions of dollars on the line, and I later led meetings with the client’s ML Logistics team to walk them through my code.

The particularly strong point was that if there were any questions, I was able to point to every single piece of equipment transiting in the simulation, so any objection or question could be scrutinized. This was also done all on my own “Pre-AI” since AIs were not capable of helping at this stage.

I started as a Business Intelligence analyst on the CEO’s team at the SQDC. Originally, the work was mostly understanding business systems, doing various analyses to identify ways to improve. However, I eventually took the initiative to start building a predictive platform that completely changes how the go-to-market cycle was done. The goals were both to automate as much as can be automated while leveraging the data we collect to make better informed decisions. Eventually, this became my core endeavor, and is becoming the backbone of the SQDC’s new predictive systems and ambitions.

The biggest thing I’ve taken from this job is a realization: the hardest part was never building the fancy model. It was designing a workflow that meaningfully changes how teams operate day to day, and managing that change so everyone feels comfortable navigating the new workflow and trusting systems they may not understand.

I’m pursuing journalism to get better at getting out there on the internet and risk failing publicly while learning how to create content. These are skills I’ll need to become the person I want to be, so I put myself in an environment that forces me to learn them. Doing this full-time degree on top of my full-time job at SQDC.

Building something to help restaurant managers. I’m always thinking about ways to make things better.

This September, I start an MBA at the University of Cambridge, as part of Jesus College and on a distinction scholarship. I’m looking forward to studying somewhere new, trying new things, and meeting all kinds of interesting people. I’m staying open to opportunities all over the world!