At an international science competition, the official business is solving the problem in front of you. The unofficial business happens while everyone waits for the results. Vladimir Novakovski has spent much of his working life thinking about the second part. A roomful of bright strangers can turn out to be more valuable than the medal. The difficulty is arranging the room.
In 2001, Novakovski represented the United States at the International Physics Olympiad in Antalya, Turkey, and came home with silver. The selection process had started with more than 1,100 students. The international tests required five hours of theoretical work and five hours in the laboratory. This was a fairly demanding way to spend a summer by the sea.
He also won silver at that year's International Olympiad in Informatics. A contemporary computing newsletter identified him as a student at Thomas Jefferson High School for Science and Technology in Virginia. Physics and programming demanded different kinds of work, but both rewarded an ability to turn a messy question into something precise enough to solve.
The tidy version of his biography would stop here, award the precocious teenager his credentials, and hurry him toward Harvard. His later career gives those early competitions another significance. They put technically ambitious young people in contact before employers, investors, and professional titles had sorted them into their respective queues. The talent was there before the business cards. What came later depended partly on who stayed in touch.
Two international silver medals. Two different ways to solve a difficult problem.
Harvard on a shorter clock
Novakovski entered Harvard at 16 and finished at 18. He explained the arithmetic years later: advanced-placement credit covered a year, and a heavier course load saved another semester. Asked whether finishing quickly had been a competition, he described it as one with himself. No rival undergraduate was apparently required.
His degree was an AB with honors. He also began doctoral work in economics as a Presidential Scholar before leaving academia. The sequence matters because the familiar image of a programming medalist does not quite contain him. Economics gave him another language for thinking about incentives, markets, and the consequences of other people's choices.
In 2004, he joined Citadel as a quantitative research associate. His subsequent account of the recruitment is pleasingly less theatrical than a legend about a billionaire discovering a teenager. He got to know the team first. Ken Griffin became involved toward the end of the process, discussing the firm's plans and the fit. Even the prodigy story, on inspection, involved colleagues, conversations, and recruitment.
Graham Capital followed in 2005. There he became a director and portfolio manager, remaining until 2012. Building a trading operation meant working through practical questions about data and systems as well as finding strategies. A promising calculation had to survive contact with an actual market. That is a useful education for anyone later tempted to confuse an elegant algorithm with a business.
The colleague who became a cofounder
His next jobs moved the same technical habits into different settings: machine learning at Quora, engineering leadership at Addepar, then the chief technology officer role at Euclid Analytics. Quora dealt in questions and answers. Addepar organized financial information. Euclid analyzed activity in physical places. Different inputs, different customers, and repeated demands to make data useful.
Euclid announced his appointment in October 2015. Novakovski spoke about the company's distinctive dataset and the prospect of turning observations from the physical world into useful analytics. He had already worked in finance and online information; now the questions included what happened inside stores. His career was accumulating domains rather than settling into one.
Addepar also supplied a relationship that outlasted the job. Scott Wu worked there as a software engineer while Novakovski was vice president of engineering. Wu later recalled concentrating on performance and optimization, alongside an unexpectedly large amount of recruiting. The people doing the hiring were themselves young enough to make the standard recruitment script look a little awkward.
In summer 2017, Wu moved to San Francisco to start a company with Novakovski. They worked from South Park Commons. The early idea, Elliot Technologies, concerned scheduling and deciding whom to catch up with. Through successive changes it became more professionally focused, then more interested in meeting new people. Lunchclub emerged through those adjustments. Hayley Leibson was also a cofounder. The founding team joined technical matching with the work of building a community.
A meeting, rather than another directory
Lunchclub proposed a surprisingly specific use for artificial intelligence: choose someone worth meeting, then help the meeting happen. Members supplied interests and objectives. The system used those signals, together with public information, to make introductions. Feedback from the encounter could inform later matches. The basic unit of value was a conversation that someone actually wanted to have.
That changes the social choreography. A directory leaves the user to identify a person, find an opening, and compose an approach sufficiently persuasive to receive a reply. Lunchclub offered to take on the selection. It was an appealing assignment for software, provided the software understood more than a job title.
A founder looking for a cofounder and an investor looking for companies could each have a good reason to join, but their useful next meetings would differ. The service had to account for intent. A large professional network might contain the right person somewhere; Lunchclub's proposition was that somewhere could become an appointment.
By March 2021, Novakovski put the objective plainly: “Our goal is to make meetings more relevant.” Mobile apps brought video conversations into the product itself. He said the service was approaching half a million users and its millionth match. Those were figures from that moment, when the company was trying to turn a promising introduction service into a regular professional habit.
“Our goal is to make meetings more relevant.”
Vladimir Novakovski · March 2021
When the lunch left Lunchclub
In March 2020, Lunchclub added video meetings. A service organized around in-person introductions could now connect people without requiring a shared café or city. The name retained its appetite; the product no longer required a table. That change enlarged the practical range of every match.
The growth was substantial. In September 2020, Lightspeed reported that monthly active users had increased fourfold in the six months to March, followed by another eightfold increase after the move to video. These were historical growth figures published by an investor. They describe the acceleration around that transition, rather than today's size of the service.
The same investor offered a small, almost comic example of the experience. Nicole Quinn was matched with a female founder she did not know. Their conversation revealed that they had grown up in the same English village and attended the same small school, a year apart. Neither profile contained that detail. It was a coincidence discovered by talking, which is precisely the sort of thing a profile cannot finish for you.
Money followed the introductions. Lunchclub's 2020 Series A raised $24.2 million, with Lightspeed and Coatue involved, at a reported valuation above $100 million. The company had attracted recognizable technology investors, but a recognizable investor list was only part of the test. People still had to accept the next match and return for another conversation.
The ambition traveled, too. In October 2021, Lunchclub announced its India launch with a target of 100,000 users there by year-end. Partnerships included WeWork and T-Hub. A target is a statement of intent, and the geographic expansion showed how much that intent had grown beyond arranging coffee in a few technology cities.
The awkward question after the good meeting
Growth and financing did not settle the business model. In his later recollection, Wu described a service that had made millions of meetings and had users who loved taking them. He also said it became difficult to grow and monetize. He left in June 2022. That account gives the Lunchclub chapter an ending more complicated than its funding announcements.
The tension is easy to understand. A good conversation can be meaningful without producing an immediate transaction. Recruiting, investing, mentoring, and exchanging ideas have different timelines and different people willing to pay. A service that brings them together must decide which value it can charge for without making the introduction feel like a sales appointment.
For Novakovski, the next move brought his earlier work back into view. In 2022, the business began its turn toward Lighter. He later described the transition as a pivot within the same corporate entity. Most of the engineering team stayed, and competing ideas were tested internally. A company built around introductions was asking its own people what they should build together next.
- 2004Trading systemsCitadel, then Graham Capital
- 2017Professional introductionsLunchclub
- 2022Verifiable marketsLighter
Back to the order book
Lighter returned him to trading, with a different technical proposition. Its decentralized exchange is built on Ethereum, using zero-knowledge proofs to make order matching and liquidations verifiable. These are consequential parts of an exchange: the rules determine how orders interact and how positions are closed when collateral becomes insufficient. Making those processes checkable is central to its pitch.
Speed still matters. So do costs and the ability to connect with other financial applications. His public conversations about Lighter keep returning to those competing demands. The problem is familiar to anyone who has built trading systems: customers experience the whole arrangement, including the delay and the bill, rather than admiring one elegant piece of it.
In November 2025, Lighter announced $68 million in funding led by Founders Fund and Ribbit Capital, with Haun Ventures and Robinhood participating. The reported valuation was about $1.5 billion. Novakovski declined to confirm that valuation. His stated ambition extended beyond a single trading venue: an infrastructure layer that could verify whether financial activity happened fairly, correctly, and transparently.
By 2026, his work also included membership of the Commodity Futures Trading Commission's Innovation Advisory Committee. In June, he discussed institutional adoption and the route to a US offering; in July, he spoke publicly about Lighter's expanded relationship with Robinhood. The conversation had widened from building an exchange to connecting it with the financial businesses people already used.
The distance from Lunchclub to Lighter is considerable. People cannot be matched like buy and sell orders, and a conversation cannot be audited into being interesting. Yet both projects ask software to improve an encounter between two sides. Novakovski's own career supplies a less programmable version of the same idea: a competition leads to colleagues, a colleague becomes a cofounder, and an introduction leaves consequences long after everyone has gone home.