Co-founder & CEO, ScoreData
Vas Bhandarkar
He left the physics of particles for the physics of prediction - and never looked back.
Vas Bhandarkar // built for the long game
He left the physics of particles for the physics of prediction - and never looked back.
In a Palo Alto office at 950 Page Mill Road, Vas Bhandarkar runs ScoreData - a company built on a quiet bet that the hard part of machine learning isn't building a model. It's keeping it honest after it ships.
ScoreData's platform, ScoreFast, builds predictive models for insurers, banks, and call centers, then watches the data feeding those models so they don't silently rot in production. The promise is unfashionably practical: take a model from weeks-or-months to days-or-hours, and let a business user, not just a battalion of data scientists, run the thing.
That framing is the whole worldview. "The world is moving from 'proprietary' tools for data scientists to Open Source tools and solutions for business users," Bhandarkar says. He has spent the better part of a decade trying to make that sentence come true in the least glamorous corners of enterprise software - claims, risk, debt-collection routing, insurance product recommendations.
In early 2024 he added a strategic-advisor seat at the InsurTechNY Fund to a resume already thick with operating roles, and late that year published a blunt note to corporate boards titled "The AI Imperative: Why Boards Must Take Bold Action." For someone who has watched four exits, the urgency is earned, not borrowed.
What sits underneath the product is a contrarian read on where the money in machine learning actually hides. The headlines go to the model - the clever architecture, the leaderboard score. Bhandarkar keeps pointing at the part that comes after: the slow decay of a model once real-world data starts drifting away from the data it was trained on. A fraud model trained last year doesn't know about this year's scams. A risk model built before a rate change misreads the new normal. The interesting engineering, in his telling, is the maintenance, not the launch.
The first credential is the surprising one: a master's in physics from IIT Bombay. Then a master's in computer science from Colorado State. Then, years later, executive education at Stanford's business school and MBA coursework at Santa Clara. Most people pick a lane. Bhandarkar kept collecting operating manuals for different machines - the universe, the computer, the company.
His early jobs read like a tour of computing's cathedrals: Bell Laboratories, Digital Equipment Corporation, Apple. But the pattern that defines him started at Remedy Corporation, an early enterprise-software company where he was an early employee before it went public on Nasdaq. He did it again at Selectica as founding VP of marketing, riding a second Nasdaq IPO in 2000.
The same year, as chairman and CEO of Unimobile, he lived the kind of growth curve founders frame on the wall: a wireless messaging platform that went from 5,000 users to 1.2 million worldwide in twelve months, then sold to Electronics for Imaging. At Cellmania he ran business development before the company was scooped up by Research In Motion, the company behind BlackBerry. And from 2004 to 2012 he sat on the board of GlobalLogic - the engineering-services firm that grew roughly 30x during his tenure and later sold to Apax Partners.
Along the way he built offshore engineering centers in Pune, Bangalore, and Delhi, and closed early partnerships with names like Ascend, BMW, and HP. He stayed wired into the ecosystem too - an active member of TiE, the global Indian-entrepreneur network, and a supporter of IIT Bombay's SINE incubator.
Then came Caralta in 2012 and, two years later, ScoreData. The founding thesis was almost economic: cheaper compute, memory, and bandwidth meant the old AI techniques could finally be used in new ways inside the enterprise. The technology wasn't new. The price of running it, suddenly, was.
Read the arc back to front and a pattern shows up. Remedy and Selectica taught him what it feels like to be early at something that becomes large. Unimobile taught him velocity - what 240x user growth in a year does to an organization, and how to sell at the top of the curve rather than the bottom. Cellmania and GlobalLogic taught him the patient version of value creation, the kind measured in board meetings and multi-year revenue charts. ScoreData is where all three lessons get spent at once: be early to a shift, move fast on it, and build something durable enough to outlast the hype cycle that surrounds the word "AI."
For most of machine learning's modern history, the model was a temple and the data scientist was the priest. You needed a rare, expensive specialist to translate a business question into math and back again. Bhandarkar's bet runs the other direction. If the tools move from proprietary black boxes to open-source frameworks, and if the platform handles the plumbing, then the person who understands the claim, the loan, or the customer can drive the model directly.
That is not a small philosophical claim dressed up as a product. It changes who holds power inside a company. It shrinks the team you need, lowers the cost of being wrong, and shortens the distance between a question and an answer. It also fits an entrepreneur who has spent a career closing the gap between deep technology and the people who actually buy it - the marketer's instinct grafted onto the physicist's training.
His message to boards in late 2024 carried the same spine. AI, he argued, is not a line item to delegate downward and forget. It is a strategic posture that has to be set at the top, with intent, before competitors set it for you. Bold action over cautious drift. Coming from someone who has been early four or five times, it reads less like a slogan and more like a habit.
Pre-built models for consumer analytics, risk, and claims compress model-to-market timelines from weeks or months down to days or hours.
The platform refreshes models at runtime using internal data, third-party sources, and regulatory information - so they keep up instead of going stale.
Built cloud-first and customized for insurance, it is designed to cut the total cost of ownership and the size of the team you need to run it.
One financial-services client improved cross-sell penetration across 500,000+ customers by 200% within two quarters.
Version control and central oversight mean less duplication, more collaboration, and faster diagnosis when a model starts to drift.
Claims, risk, and insurance product recommendations - the unglamorous decisions where small accuracy gains pay for themselves.