The pitch is deceptively simple: Before artificial intelligence writes your tax return or recommends a cancer treatment, shouldn't someone be able to prove the machine got it right?
That question landed Pramaana Labs a $27 million seed round announced on June 17, 2026, led by Khosla Ventures with backing from Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound. The San Francisco startup is building what it calls formal verification tools for AI—mathematical proof systems designed to catch mistakes before they spiral into lawsuits, overdoses, or IRS audits.
It's an unusually large seed for a company that, according to LinkedIn as of mid-June 2026, employed somewhere between 11 and 50 people. The company's website lists additional investors—Founders Future and the Perplexity Fund among them—names that appear on those investors' portfolio pages but were conspicuously absent from the initial press materials. Perhaps the cap table got complicated. Or perhaps not everyone wanted top billing.
Verification as a Service
Pramaana describes itself as a "frontier AI lab for verifiable intelligence," which is the sort of phrase that sounds impressive until you ask what it actually means. The core idea, though, is straightforward enough: Large language models hallucinate. They sound confident when they're wrong. And in domains like drug interactions or multi-state tax compliance, sounding confident while being wrong can get expensive fast.
The company's flagship product, the Domain Formalizer, takes dense regulatory text—think IRS code, FDA guidelines, legal statutes—and converts it into machine-checkable logic. It's not just parsing documents; it's translating human rules into a formal language that can be verified using proof systems borrowed from mathematics and computer science.
Under the hood, Pramaana layers the Lean proof assistant over large language models. Each output comes with what the company calls "proof artifacts"—essentially, a mathematical trail showing how the AI arrived at its answer. The result is auditable, defensible, and (in theory) far less prone to the kind of hallucinatory nonsense that has plagued generative AI since its commercial debut.
The three-step process, as outlined on Pramaana's website: formalize domain knowledge, constrain inference within logical boundaries, then verify outputs through both self-consistency checks and formal proof validation. Examples include multi-state sales tax calculations and systems that detect conflicts buried deep in regulatory frameworks.
Target markets span six verticals: tax and statutory reasoning, legal and regulatory compliance, healthcare and clinical safety, scientific research, financial modeling, and misinformation detection. It's an ambitious list. Whether any startup can credibly tackle all six at once is another question.
The Founders: Google Alumni with a Proof Problem
Three co-founders with serious technical credentials are steering the ship. Ranjan Rajagopalan, the CEO, graduated from IIT Madras and previously co-founded a startup called Astra. Before that, he worked as a software engineer at Google and Graviton, focusing on high-performance architecture and machine learning pipelines—the kind of résumé that opens doors in Silicon Valley.
Krishnan Raghavan, the CTO, built Glean's India-based search team and spent years as a Staff Software Engineer at Google. His LinkedIn profile lists a computer science degree from IIT Madras, making him and Rajagopalan part of the well-worn pipeline that funnels top Indian engineering talent into the Bay Area.
Then there's Sanjay Ganapathy Subramaniam, Chief Scientist. He came from Google DeepMind, where he architected the tool-use system for Gemini and led post-training work on what the company calls "frontier capabilities." That's insider speak for the cutting-edge stuff that makes headlines—and occasionally regulatory nightmares.
The team has also recruited Danny Werfel, who left his post as IRS Commissioner in January 2025, to advise on tax law formalization. (Nothing says "we're serious about compliance" like hiring the guy who used to run the IRS.) Academic collaborators from IIT Delhi, IIT Madras, and UC Berkeley are working on cybersecurity and drug discovery applications, though details remain vague.
Riding the Wave—or Creating It?
Timing here matters. Pramaana's round arrives as formal verification starts creeping out of academic labs and into the AI mainstream. Mistral AI released Leanstral, an open-source proof agent for Lean 4, in March. DeepSeek shipped Prover-V2 for formal math proofs the year prior. And Axios reported in late May that a startup called Axiom had successfully gotten machine-checkable proofs published in peer-reviewed journals—a first, and a signal that the field might be maturing faster than expected.
Just a week before the funding announcement, Pramaana hosted something called The Verification Summit in San Francisco. The June event featured a fireside chat between Vinod Khosla and Rajagopalan, optimistically titled "The verification thesis," along with research presentations from UC Berkeley, Stanford, NVIDIA, and Microsoft Research. It closed with the unveiling of the Pramaana Research Program, though specifics on what that entails remain thin.
There's some murkiness around when exactly the company got started. LinkedIn lists the founding year as 2025, while Founders Future lists the creation as 2026. India corporate records show that RKS Pramaana Labs India Pvt Ltd was incorporated in Bengaluru on September 12, 2025, with Rajagopalan and Raghavan named as directors. Best guess? Operations began in late 2025, followed by a more formal U.S. launch the following year. Startups rarely get their own origin stories straight.
The Infrastructure Play No One's Talking About
Pramaana hasn't said much about how it plans to spend the $27 million, but the positioning offers clues. This isn't a product company in the traditional sense—it's building picks and shovels for an AI gold rush that may or may not materialize. The website calls for research fellows and partners, signaling an emphasis on expanding academic ties and industry collaborations rather than racing to a consumer-facing launch.
The underlying wager is clean: As AI systems infiltrate environments where mistakes aren't just embarrassing but legally actionable—tax filings, pharmaceutical protocols, contract review—someone will need to build the infrastructure that guarantees correctness. Or at least guarantees the appearance of correctness, which in regulatory contexts might be nearly as valuable.
Whether Pramaana becomes that someone depends on variables the founders can't fully control. Regulators might mandate formal verification before they're ready to deliver it at scale. Competitors with deeper pockets could move faster. Or the market might decide that "good enough" AI, backed by insurance policies and liability waivers, is cheaper than provably correct AI.
For now, though, Khosla Ventures and its co-investors are betting that when the music stops, the companies that can prove their AI did the math right will be the ones left standing. It's a persuasive thesis. Whether it's a $27 million thesis remains to be seen.
