He spent years inside Google Maps learning how quickly the physical world drifts out of date. Now his startup VOYGR is betting that AI agents need a map that knows more than a name and a rating.
Ask a map app about a restaurant and it hands you a tidy little card. Four-point-two stars. Open till ten. A phone number, maybe a photo of the front door. What it does not tell you is that the chef left in March, the wait times doubled, and the regulars quietly migrated two blocks over to the new place. The card looks fine. The card is wrong. And for fifteen years, Vlad Baskakov has been standing close enough to that card to see exactly how it lies.
Baskakov is the co-founder and CEO of VOYGR, a San Francisco company in Y Combinator's Winter 2026 batch. The pitch is deceptively small: a better maps API. The problem underneath it is not small at all. Every day, the physical world changes in millions of tiny ways that no one bothers to write down. Places open, close, rebrand, move, raise prices, change hours. None of it arrives as a clean signal. As Baskakov puts it, the physical world does not emit structured "I closed" events. Someone has to go and find out.
For most of the internet's history, that someone was a human squinting at a screen, deciding whether the listing looked plausible. Now the someone is increasingly a machine - an AI assistant booking a table, an agent planning a route, a model answering a question that quietly depends on whether a coffee shop still exists. And the machines, it turns out, are extremely confident about places that closed a year ago.
Good founders rarely invent their problems. More often they have already lived inside them for years, in different disguises, until the shape of the thing becomes impossible to ignore. Baskakov's route to VOYGR reads like a slow accumulation of evidence.
He studied computer science and business together at the Higher School of Economics in Moscow - an early tell for the product-and-engineering hybrid he would become. From there came McKinsey, then a Harvard MBA, then a stretch in two of the hardest markets a map ever has to serve: ridesharing at Gett and travel at OYO. Both businesses live or die on whether a location is accurate at the exact moment a person needs it. A pin in the wrong spot is a driver who can't find you. A stale address is a traveler stranded at the wrong door.
Then Google Maps, where he led product strategy and go-to-market for places and local advertising, worked on the Maps API products developers build on, and touched the data-sharing between Maps and Gemini. This is the vantage point that matters. From inside the biggest map on earth, he could see something that looks obvious only in hindsight: the industry treated place data as a feature to be shipped, not as infrastructure to be maintained. Freshness was everyone's assumption and no one's job.
Here is the number that reframes everything. Somewhere between a quarter and a third of all places churn every single year. Restaurants close, shops move, hours shift, businesses change hands and names. That is not an edge case buried in a spreadsheet. That is the ground moving under every app, every search result, and every AI answer that touches the real world.
And the real world is stubbornly quiet about it. A website can ping you when it changes. A database can log a write. A shuttered diner does nothing at all - the lights just stay off. To know it closed, you have to actively look: read the news, watch the reviews sour, notice the social account go silent, cross-reference the signals until you are confident. That detection problem is the technical heart of VOYGR, and it is why Baskakov keeps returning to the phrase that has become the company's spine.
Stack those figures together and the stakes come into focus. Local context is not a niche. It sits under a huge share of what people ask software to do, and an even larger share of what they will soon ask agents to do on their behalf. If the layer underneath is stale, everything built on top inherits the error - confidently, and at scale.
The second half of Baskakov's thesis is about depth, not just freshness. A standard map listing carries maybe ten to fifteen attributes: name, address, category, hours, a rating, a phone number. That was plenty when a human was doing the browsing and filling in the gaps with judgment. It falls apart the moment a machine tries to answer something specific.
VOYGR's framing is a "queryable place profile" that expands far past those fifteen fields into something closer to open-ended. The company likes to demonstrate the range with a deliberately absurd example - it can surface which sauna a celebrity like Justin Bieber is known to frequent - precisely because no traditional map would ever carry a detail that strange. The point is not the sauna. The point is that once you stop capping attributes at fifteen, you can start answering the questions people actually type into a chatbot.
Consider the query VOYGR uses to make the idea land: "specialty coffee shops in SF with Wi-Fi, popular with YC founders." No collection of fifteen tidy fields answers that. It needs semantics, fresh web context, and a sense of the texture of a place. That is the difference the company draws between what maps do and what it wants to do.
Baskakov did not build this alone, and the split is telling. His co-founder and CTO, Yarik Markov, spent more than a decade leading machine learning, search, and infrastructure teams at Apple, Google, and Meta - the kind of production systems that quietly serve hundreds of millions of people. One founder knows the market, the customer, and the go-to-market motion cold. The other has spent his career building the plumbing that makes something like continuous place detection actually work at scale.
It is a clean division of the two hardest questions a company like this faces. Who needs this, and how do you sell it? And how do you actually detect a changing world without drowning in noise? VOYGR already reports processing tens of thousands of places daily for enterprise customers, with validation precision cited as high as 99.62 percent - the sort of figure that only matters if you have someone who obsesses over the last fraction of a percent.
The timing argument is straightforward. Maps used to be browsed. A person opened an app, looked around, made a decision. Now maps are increasingly read by models - agents that use place data to reason and then do something with it: book, route, recommend, transact. Google Places has been a multi-billion-dollar business for years, but Baskakov's wager is that agents change its nature entirely, turning a passive lookup into an active, transactional layer for local commerce. Whoever supplies the understanding underneath that layer is in a very interesting position.
The "why him" is the quieter part, and maybe the stronger one. This is a founder who watched place data go stale from the demand side at Gett and OYO, from the supply side at Google, and from the strategy side of the ads business that monetizes it. VOYGR is not a clever idea he stumbled into. It is the thing he could not stop noticing, built by the person best positioned to notice it. The company's whole promise fits into three words that double as a mission statement and a shrug at the incumbents.
There is a version of the coming years where AI agents quietly run a large share of our small daily errands, and the difference between a good experience and a maddening one comes down to something invisible: whether the map underneath knew the truth. Baskakov is building for that version. He has been building toward it, in one form or another, for fifteen years. The card on your screen still says open till ten. He is one of the few people who has spent a career refusing to take it at face value.