Two founders who failed at groceries are now backed by Nvidia to reinvent how humans collaborate with math — the kind of unsexy infrastructure play that actually matters when AI needs to do more than generate text.
The Summary
- Corca raised $7.8 million from Nvidia's NVentures, NEA, Bloomberg Beta, and Daft Capital to build collaborative math software that positions itself as "Cursor for math"
- Cofounders Oleg Shevlyagin and Anton Gladkoborodov previously ran competing NYC grocery delivery startups that both failed after acquisition deals fell through
- The workspace lets users write equations, run calculations, get AI assistance, and collaborate in real time — targeting the gap left by 30-year-old tools like MATLAB and LaTeX
The Signal
Corca is attacking a problem that doesn't sound exciting until you realize every AI lab, quant fund, and engineering team hits it daily: sharing math is still brutally manual. MATLAB costs thousands per license and assumes you're working alone. LaTeX requires learning a markup language just to write a fraction. Teams end up screenshotting equations or, worse, photographing handwritten notes and dropping them in Slack.
Shevlyagin calls it "one of those sleeping categories where nothing's going on for more than 30 years." He's right. The last major innovation in mathematical notation software was Mathematica in 1988. Since then, we've gotten incremental updates to the same paradigm: you the human, writing code to tell the computer what math you want to see.
"It's like one of those sleeping categories where nothing's going on for more than 30 years."
Corca flips that. Instead of learning LaTeX syntax, you write math naturally and the AI handles formatting. Instead of emailing Jupyter notebooks back and forth, your team works in the same workspace, real-time, like Figma but for differential equations. The "Cursor for math" framing is smart marketing, but the actual play is broader: they're building the collaboration layer for quantitative work that Web2 tools never bothered with because the market seemed too technical, too niche.
Nvidia's involvement tells you where this goes. NVentures doesn't write checks to nice productivity tools. They invest in infrastructure for the next computing paradigm. If AI agents are going to do serious technical work, they need to manipulate mathematical objects, not just summarize papers about them. Corca is building the interface where humans and agents can work on the same equation at the same time.
Key infrastructure needs for agentic math work:
- Real-time collaboration (humans checking agent work, agents extending human proofs)
- Version control for mathematical objects (like Git for equations, not just code)
- AI-native formatting (agents shouldn't need to learn LaTeX any more than humans should)
The founder story is also worth noting. Shevlyagin and Gladkoborodov ran rival grocery delivery companies that both collapsed. They could have stayed rivals, blamed each other, moved to different cities. Instead they started meeting to discuss physics and got annoyed enough at the tools to build something new. That's the kind of founder origin that matters: they're users first, company-builders second. They founded Corca in 2023 with 12 employees and have already attracted users, though the article doesn't specify numbers.
The Implication
Watch for Corca to become essential infrastructure in AI research labs over the next 12 months. If agents are going to contribute to technical work, not just summarize it, they need native-quality math manipulation. That's not a feature Microsoft will bolt onto Office. It's a new category.
For individual researchers and quant teams: this is the collaboration wedge. You adopt it because your team is tired of LaTeX hell. You keep it because the AI assistant gets good enough to check your work, then extend it. The same pattern Cursor followed in code is about to play out in math.