Pramaana Labs has successfully raised $27 million in seed funding, led by Khosla Ventures, with additional participation from Accel, Boldcap, Nexus Venture Partners, Premji Invest, and Unbound. The startup aims to integrate formal verification methods into artificial intelligence (AI) systems to enhance their reliability, particularly in critical sectors such as law, drug discovery, and tax preparation.
Formal verification involves using mathematical techniques to prove or disprove the correctness of algorithms, ensuring that systems operate without errors. By applying these methods to AI, Pramaana seeks to mitigate issues like hallucinations and inaccuracies that can arise in complex AI models. This approach is especially vital in fields where errors can have significant consequences.
Pramaana’s strategy combines large language models (LLMs) with a deterministic verification layer. This setup allows the AI to handle natural language queries and complex tasks while ensuring that its outputs are accurate and reliable. The company utilizes the open-source LEAN programming language, known for its applications in verifying mathematical proofs, to build this verification layer.
For each application area, Pramaana collaborates with domain experts to develop tailored formal verification systems. In the realm of tax law, for instance, the company is working with former IRS commissioner Danny Werfel. Additionally, professors from institutions such as IIT Delhi, IIT Madras, and UC Berkeley are contributing to the development of systems for cybersecurity and drug discovery.
The integration of formal verification into AI represents a significant advancement in ensuring the dependability of AI applications in high-stakes environments. As AI continues to permeate various industries, the demand for robust and error-free systems becomes increasingly critical. Pramaana Labs’ innovative approach addresses this need by combining the flexibility of LLMs with the precision of formal verification, setting a new standard for AI reliability.
