AI hallucinations are a known problem, but Pramaana Labs is trying to solve it with math — specifically, the same kind of formal verification used to prove mathematical theorems. The startup has raised a $27 million seed round led by Khosla Ventures, with participation from Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound. The money will go toward building a deterministic verification layer for industries that can't afford mistakes: law, drug development, tax filing, and more.
Pramaana's system still runs a traditional large language model underneath — that's where the flexibility comes from. But on top, it adds a verification mechanism inspired by the open-source LEAN programming language, which is used to formally verify mathematical proofs. The idea is to encode industry-specific rules into this formal system, so the AI's outputs get checked before they're trusted.
To build LEAN-style verification for different verticals, Pramaana is working with domain experts. In tax, that means collaborating with former IRS commissioner Danny Werfel. In cybersecurity and drug discovery, professors from IIT Delhi, IIT Madras, and UC Berkeley are leading the effort.