Interning at Instabase: The 10 Million Dollar Ideas

Aug 30,2019

On my third day of work at Instabase, our CEO Anant plopped down next to me, dumped his laptop on the desk, and warmly declared, “Hello, neighbor.” I knew that working at a start-up necessitated constant desk reorganization, but I really didn’t expect a seat next to the big boss. And I certainly could not have predicted his next words to me: “What’s the ten million dollar idea that you’re working on?”

Despite that responsibility being terrifying, it wasn’t much of an exaggeration. Instabase has dozens of huge customers that rely on the platform to make structure out of chaos for their hundreds upon thousands of documents. I was originally tasked with revamping our core OCR for Driver’s Licenses (Emerson), and as OCR is our bread and butter, getting quality improvements became the holy grail of my internship. I was able to toe the line between an engineer and a data scientist as I refactored old, academic code that I might have written for a school project, while still making algorithmic changes to the underlying model. A month into my internship, we shipped Emerson v2 to a customer, and it was merged into production soon after.

Emerson (L) vs. Google’s API (R) — decent results, especially for fields that someone would try to extract (names, DOBs, expiry dates, etc.)

While I could have spent the rest of the summer making the next set of improvements to Emerson, the head of my team, Ted, had other plans. He recognized that my strength came in creative model design and wanted to move me onto a more experimental project where we could make more headway through my internship. I appreciated the agility and thoughtfulness about what we put on people on. Labor is fundamentally most limited resource for a start-up of our size and growth, and everyone works more productively when operating in a niche.

After Anant was satisfied with the first-round of Emerson improvements, he then went after the next ten million dollars: “come up with a robust OCR for handwriting.” This is what I spent the latter half of the summer doing, and it’s one of my favorite projects that I have worked on over the past few years. It forced me to scope a tool from the ground-up in tandem with customer development, and exercise product and design skills. It also opened my eyes to what I believe is the future of AI: contextual learning. Nowadays, it’s not impressive to have a decent generalized model for a solved problem; big tech giants have had OCR models for handwriting for years. But enterprise companies are just now realizing that these models fail in specific cases where context can be leveraged⁠ — for example, if I’m told that a certain picture of handwriting is supposed to be a US credit card number, I automatically know that it’s a 16-digit number. This valuable information cannot be embedded into an API call to Google Vision, but it can affect the output that Instabase produces for its custom OCR tools. The automation economy is here, and “good enough” doesn’t cut it anymore⁠ — the world wants specific and precise solutions that fit their business needs.

And maybe I’m drinking the Kool-Aid when I express this sentiment, but I went to school in a tech bubble. Arguably, MIT isn’t as crazy as the Silicon Valley tech bubble, but it’s bubble all the same. I’ve heard my fair share of crazy corporate stories and think I’m cynical enough not to buy into every ridiculous start-up slogan I hear. But if there’s one thing that has been drilled into me time and time again in every industry experience and internship I’ve had, it is that the rest of the world has not quite yet caught up to the crazy technology that I play with every day, and introducing them to it would make their lives a lot easier. People are usually resistant to change, but there’s less friction here at Instabase. Change is inevitable, I suppose, from us switching offices halfway through my internship, or my newly-acquired caffeine addiction as a result of the iced tea that’s stocked in our fridge. New Instabase hires are excited to modify internal processes. We reprioritize often, and that’s healthy.

As a part of my internship, I also built a benchmarking tool to test Emerson against its competitors and give quantitative insight as to where we needed improvement the most. There has been a palpable shift at the company since I joined, and we’re finally reaching the stage where everyday decision making is based on rigorous analysis rather than gut instinct. I was shocked that wasn’t initially the case, and as an engineering student, I’ve been conditioned to back up calls with facts and figures. But more and more I have realized that companies, and especially Instabase, are not solely built on charts and slides with dry statistics.

Companies are fundamentally driven by missions. Instabase was created with a vision, and while it may may morph over time, these visions shape our culture, people, and product. This isn’t to say, of course, that we should stop fostering a data-driven environment, but rather that the company is inspired because they believe that businesses can work smarter, and not just because the data points to this as well. It’s rare to see such an influential company of our size, and it’s been a pleasure to see the company grow even in the few months that I have been here. I chose to do an internship at Instabase because it tackles big data problems in particular verticals and really changes the game for all of our customers; I can’t wait to hear about the ten million dollar idea that the next batch of interns will get to work on.