As organizations continue to face challenges in transforming AI pilot projects into dependable business solutions, reliability has emerged as one of the industry’s most pressing concerns. A new startup believes it has found an answer by combining the rigor of mathematical formalization with the flexibility of modern artificial intelligence, bringing together one of computer science’s most trusted methodologies and one of its most unpredictable technologies.
On Wednesday, Pramaana Labs announced that it had raised $27 million in seed funding in a round led by Khosla Ventures. Additional investors participating in the financing include Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound.
The company plans to focus on high-stakes industries such as legal services, drug discovery, and tax preparation, where even minor mistakes can result in significant financial, regulatory, or personal consequences. In these sectors, deploying AI systems requires far stronger safeguards against hallucinations, inaccuracies, and unreliable outputs than those currently available in most generative AI products. According to Pramaana co-founder and CEO Ranjan Rajagopalan, these industries are particularly well suited to a more structured and formalized approach.
“It’s similar to mathematics because there are numerous rules and constraints that must be followed,” Rajagopalan told TechCrunch while discussing the complexity of tax regulations. “Once those rules are properly codified, the reasoning built on top of them begins to behave in a much more deterministic way.”
Pramaana’s platform continues to rely on conventional large language models, allowing it to maintain the flexibility required to understand natural language, answer complex questions, and solve problems that traditional software systems struggle to address. However, the company adds an additional deterministic verification layer above the LLM, ensuring that the model’s conclusions and reasoning processes can be validated before being trusted.
The concept of combining large language models with deterministic verification mechanisms has gained popularity across the AI industry. What differentiates Pramaana is its use of formal verification techniques inspired by LEAN, the open-source programming language widely used for verifying mathematical proofs and ensuring logical correctness.
There is already evidence that this type of approach can succeed in practice. Rajagopalan points to France’s CATALA project, which has converted large portions of the country’s tax and social benefits framework into executable code. By formally representing legal and regulatory systems, projects like CATALA demonstrate how complex rule-based environments can be transformed into structures that computers can verify and execute reliably.
For each industry it serves, Pramaana intends to develop specialized formal verification frameworks modeled on LEAN. These systems will be built and supervised in collaboration with subject-matter experts to ensure accuracy and domain-specific compliance.
In the field of tax law, the company is working alongside former IRS Commissioner Danny Werfel. For cybersecurity and drug discovery initiatives, Pramaana has enlisted academic experts from IIT Delhi, IIT Madras, and UC Berkeley to help guide the development and validation of its verification systems.
Rajagopalan believes that many of society’s most difficult challenges remain unresolved not because they are impossible to solve, but because they have yet to be formally defined in a way that computers can reason about reliably.
“The world’s hardest problems are not unsolvable. They are unformalized,” Rajagopalan said. “Every field where mistakes can impact a person’s health, finances, or freedom is governed by rules.”
The next step, he argues, is ensuring those rules are systematically codified, verified, and integrated into AI systems capable of operating with far greater reliability than today’s models.