Feature Labs Brings Accessibility to the Complex World of Data Science and Machine Learning
If you’re involved with the ever-expanding world of technology, there’s no doubt you’ve heard about machine learning. Startups are coming out of stealth mode proclaiming they are utilizing machine learning for their platform/service/software/etc.
However, as the adoption rates rapidly grow, it may not come as a surprise to learn that there are some issues with teams trying to jump on the trend.
Founded in 2015, Feature Labs is a startup out of seemingly-endless MIT startup ecosystem that is offering assistance to companies that are starting to utilize this type of technology. Their name comes from a subset of data science engineering called “feature engineering.”
Feature Labs’ origins involve a lab on MIT’s campus and a connection to a massive online community
In 2014, Future Labs Co-Founder and CEO Max Kanter wanted to obtain his masters from his alma mater MIT, and two subjects he wanted to study was machine learning and applied research. His interest in these subjects stemmed from the development of a recommendation engine that he built for readers of the New York Times’ website, which used the ubiquitous technology. Kanter came across the Data to AI Lab on campus, which is also where he met his future co-founder, Dr. Kalyan Veeramachaneni, principal investigator of the lab. The two had similar backgrounds when it came to machine learning as well.
“As practicing data scientists, we were spending most of our time taking the raw data from the database and getting it ready to plug it into tools. That was all manual and was the main barrier for a lot of people to get involved with machine learning,” said Kanter. “We started looking into developing technology to help automate that missing link. That’s what enabled the process to be faster, but also to get more people involved.”
Kanter began working as a student researcher with Veeramachaneni, and they started brainstorming a way to create a tool to help solve their common issue with machine learning and data science. As part of his graduate thesis, Kanter created an algorithm that would be able to take in large amounts of data. and was able to showcase it through Kaggle, the largest online community for data scientists.
The results from the algorithm resulted in a research paper that ended up gaining traction within the data science community. It also led to various publications taking note of the power behind this technology.
In classic startup fashion, a project became a full-fledged idea for a business. Kanter and Veeramachaneni decided to take the plunge and start Feature Labs in 2015.
The extensive Feature Labs platform and what it does
The most significant problems Kanter and Veeramachaneni are looking to solve are that machine learning is difficult to understand, and that it is prone to human error. To alleviate these pain points for some companies, their platform is automating parts of the data science and machine learning process.
The platform’s most prominent trait is its ability to bring in raw data and put them into datasets for machine learning algorithms, without having someone pour over tons and tons of data.
“Imagine you are an eCommerce company, and you’re trying to predict what product a customer is going to buy next. Chances are you’re going to want to find out other products they’ve looked at, what searches they’ve made, what they’ve added to their cart, and their previous orders,” said Kanter. “Before you can input that data into a machine learning algorithm, you have to come up with explanatory variables. Traditionally, data scientists would take a look at the raw data and figure out the variables themselves.”
How Donnelly met the team...
In February 2018, Feature Labs announced its seed funding led by Flybridge Capital totaling $1.5M. It was around this time, John Donnelly came onboard the company as Chief Operating Officer.
Donnelly has spent much of his career in the Boston tech scene, with his most recent position being Senior Vice President at Crimson Hexagon and is an advisor to early-stage companies that are from the MIT ecosystem.
“I’m familiar with the Flybridge team, and I know Jeff [Bussgang] pretty well. Jeff and I had always talked about trying to find a company within the portfolio to work on together,” said Donnelly. “So, he made some introductions to his partner Chip Hazard and a few of the startups in his portfolio, and I met Max and the Feature Labs, team. I started to work with them as an advisor, but became more interested in joining full-time, building out a go-to-market strategy and finding the right machine learning projects and use cases where enterprise companies can use Feature Labs.”
“There’s a lot of companies out there selling data science platforms, but there’s not a lot that are taking the most important aspect of building accurate predictions into account,” Donnelly added.
Aside from continously updating their platform, Feature Labs is busy taking part in events and gatherings in both the MIT and Boston startup ecosystems.
“We have been part of MIT Startup Exchange as part of their STEX25. Every year, MIT picks 25 of their market-ready startups. They take the MIT startups that they feel have been developed enough and then bring them to the school’s partner companies,” said Kanter. “We’ve also been in attendance at nearly every MIT startup event, but there haven’t been a lot of collaboration with other startups. Not yet at least.”
What does the company think of the future of “feature engineering?”
“The industry is evolving. In the last six months, we’ve seen a lot of relatively conservative enterprise companies across the space try to bring more automation to their processes,” said Donnelly. “They are hiring data science teams, but as Max described earlier, it is time-consuming for someone to learn it. Feature engineering is becoming more mainstream as more and more companies try to bring automation on board.”
Images courtesy of Feature Labs