Enterprises are still figuring out how to transform AI from pilot projects into reliable, production-ready business tools, with reliability issues becoming the primary barrier to deployment. Pramaana Labs has announced the completion of a $27 million seed round, planning to add a verifiable deterministic layer to large models using formal verification methods.
This round of funding was led by Khosla Ventures, with participation from Accel, Boldcap, Nexus Venture Partners, Premji Invest, and Unbound. The company will initially focus on highly sensitive industries such as taxation, legal services, and drug development, where errors often carry high costs.
Start with high-risk industries.
Pramaana believes that industries with complex rules are better suited for encoding and verification. For example, tax regulations consist of numerous clear provisions. Once these rules are structured into executable formats, subsequent reasoning can become more deterministic, rather than relying entirely on model-generated outputs.
Ranjan Rajagopalan, Co-founder and CEO of the company, said that many high-risk industries are not beyond AI’s capability, but rather have not yet formalized their rules. By converting key rules into verifiable systems, biases and hallucinations in models during critical tasks can be reduced.
Add a verification layer to the large model
Pramaana’s approach does not abandon traditional large models. Its system still relies on conventional LLMs at the core to preserve natural language understanding and the ability to handle complex queries; built on top of this is an additional layer of deterministic verification to ensure model outputs comply with predefined rules.
This approach draws on formal verification tools. The company noted that its method references the open-source programming language LEAN, which is commonly used to verify mathematical proofs. Rajagopalan also mentioned that France’s CATALA project has converted parts of its tax and welfare systems into executable code, providing a real-world example of a similar path.
Invite domain experts to participate in modeling
For different application scenarios, Pramaana will establish corresponding formal verification systems, supervised by industry experts. In the tax domain, the company is collaborating with Danny Werfel, former Commissioner of the Internal Revenue Service of the United States. For cybersecurity and drug discovery, professors from IIT Delhi, IIT Madras, and the University of California, Berkeley are involved.
According to the company’s vision, such systems are suitable for tasks with significant health, financial, or legal implications. The goal is not to have the model make decisions independently, but rather to enable the model to reason and respond within a set of verifiable rules, thereby enhancing controllability during enterprise deployment.
- Funding amount: $27 million in seed round
- Lead Investor: Khosla Ventures
- Key scenarios: Taxation, legal, drug development