Startups & Funding
Pramaana Labs raises $27M to bring formal verification to AI
Pramaana Labs raised $27 million in seed funding to combine LLMs with formal verification tools, aiming to improve reliability in sensitive sectors like law and tax.
On Wednesday, Pramaana Labs announced a $27 million seed funding round. The investment was led by Khosla Ventures, with a syndicate of participating investors including:
- Accel
- BoldCap
- Nexus Venture Partners
- Premji Invest
- Unbound
The startup aims to address AI hallucinations and errors by introducing formal verification—the use of mathematical methods to check if a system behaves correctly. Pramaana Labs uses the LEAN programming language, an open-source programming language used to verify mathematical proofs, to build a deterministic layer on top of conventional Large Language Models (LLMs). This setup allows the system to process natural language while ensuring its outputs follow strict, verifiable rules. Rajagopalan compared the structure of these domains to mathematics, noting that they operate under a strict set of rules that must be followed. He explained that once a domain’s rules are codified, the reasoning built on top of them becomes deterministic.
Pramaana Labs will focus on sensitive verticals including law, drug discovery, and tax preparation. These are areas where being wrong can cost someone their health, money, or freedom. Rajagopalan stated that the most difficult problems in the world are not unsolvable, but rather unformalized. “Every domain where being wrong can cost someone their health, money, or freedom has rules,” said Ranjan Rajagopalan, co-founder and CEO of Pramaana Labs.
To build these verification systems, the company is partnering with domain experts and academic institutions. For its tax systems, Pramaana Labs is working with former IRS commissioner Danny Werfel. For its cybersecurity and drug discovery systems, the company is collaborating with academic partners from IIT Delhi, IIT Madras, and UC Berkeley.
This approach of formalizing complex regulatory frameworks has precedent. Rajagopalan pointed to the CATALA project in France, which formalizes the country’s tax and benefit systems into executable code. By applying similar formal verification methods to AI, Pramaana Labs plans to build custom verification systems for each of its target industries, ensuring that LLM outputs can be mathematically validated before they are deployed in high-stakes environments.
Why it matters
Pramaana Labs is attempting to solve the reliability issues of AI in sensitive verticals by combining conventional LLMs with deterministic formal verification tools. If successful, this hybrid approach could allow enterprises to deploy AI in sensitive verticals where errors carry severe consequences.