The Search Engineer Who Refused to Rebuild from Scratch
Eli Finkelshteyn's Master's thesis was on sentiment analysis in ancient Aramaic. This is not a metaphor. At Brandeis University, while earning a triple major in Math, Economics, and Computer Science before pivoting to a graduate degree in computational linguistics, he trained models on texts that predated the printing press. It was impractical, methodologically rigorous, and somewhat inexplicable - exactly the kind of thing that makes a great search engineer.
His first industry stop was FactSet Research Systems, working on natural language processing. Then Tumblr, where the job was spam and phishing detection at scale. Then Shutterstock in 2011, where the real story begins. As a search engineer there, Finkelshteyn kept bumping into the same problem: building a genuinely great product search experience requires not just a search engine, but a behavioral tracking layer, a big data cluster, and machine learning algorithms trained on shopping patterns. Most companies couldn't build all three. Most companies kept trying anyway.
He met Dan McCormick at Shutterstock - McCormick was then the company's CTO. They shared a frustration that would become a company thesis: across the entire e-commerce industry, teams were duplicating enormous engineering effort to solve the same search problem. It was like every contractor in the city building their own hammers.
Let's just do it once. We'll do it well, and then nobody ever has to do it again.
- Eli Finkelshteyn on founding ConstructorConstructor was founded in 2015. What happened next is the part that gets overlooked: they did not launch. For four years, Finkelshteyn and McCormick and a small team built the product, refined the architecture, trained the systems, and deliberately stayed out of the press. In an era defined by fast launches and faster pivots, they chose patience. The platform that eventually went to market in 2019 had four years of product development behind it before its first enterprise customer ever logged in.
That first customer was Jet.com, the Walmart-owned marketplace that was then handling thousands of queries per second. Constructor had four employees. Jet ran an A/B test on Constructor's autosuggest. The results moved the needle. A new category was established.
The pitch was straightforward and remains so: e-commerce search built with behavioral machine learning, not static Elasticsearch configurations. The distinction matters because of what Elasticsearch can't do. "You can't just build it on top of Solr or Elasticsearch," Finkelshteyn has said, "because then you eventually plateau way too early." Constructor's engine continuously learns from shopping behavior - what people click, what they buy, what they ignore - and adjusts results accordingly. The process is descriptive, not prescriptive. The system learns what shoppers want rather than following rules someone programmed about what they should want.
Personalization, in Finkelshteyn's framing, is also a privacy problem. Constructor uses collaborative filtering rather than personally identifiable information - the same technique behind "customers also bought" recommendations, but without tagging individual users. You can personalize without being creepy. He has said this directly and built the product accordingly.
The Series B raise in 2024 was oversubscribed. Sapphire Ventures led the round at a $550M valuation - triple the company's value from its 2021 Series A. Constructor wasn't actively fundraising. Investors called them.
The company went from 233% year-over-year revenue growth at its Series A announcement in September 2021 to doubling revenue in each of the three following years. The client list grew to include Sephora, Under Armour, Gap, REI, Bonobos, American Eagle, Target Australia, and Birkenstock. Client retention sits at 98.5% - a figure that reveals more about product-market fit than most marketing copy ever could.
Finkelshteyn's philosophy on leadership is equally specific. He credits his high school English teachers - Thomas Johnson, J.J. Hurley, and Joani Reese - with teaching him how to connect with an audience. He is explicit that leadership means being selfless. "If you need to be the hero, let someone else lead" is not a motivational poster aphorism in his case; it traces to something personal. In college, his stepbrother died in a car crash. Finkelshteyn has described how their relationship before that had operated like a zero-sum game. The loss changed his framework: any success he pursued would be shared with the people around him. Constructor's organizational culture - where ideas flow from everywhere and credit goes to whoever generated them - reflects that.
He openly acknowledges the ongoing difficulty of work-life balance. He is a father who prioritizes fatherhood. He describes being constantly calibrated toward the question of having the greatest impact - in work, in his team, in the broader market. The company he runs now processes more search queries in a year than most databases see in a lifetime. The ancient Aramaic models are long retired.