Pramaana's $27M Seed Brings LEAN Formal Verification to LLMs

On June 17, 2026, San Francisco-based AI reliability startup Pramaana Labs announced a $27 million seed funding round led by Khosla Ventures. The company is building a verification layer designed to convert probabilistic large language model outputs into mathematically provable statements. The funding will scale the training of formalization models and expand the company’s domain expertise research teams.

Deterministic Truth via Formal Proofs

Large language models generate text based on probability distributions, which creates an accountability gap in high-stakes environments. Pramaana approaches this limitation through Formal Verification. The platform utilizes the Lean programming language to build a deterministic layer over the generative models.

Instead of allowing a model to estimate a legal rule or tax deduction, Pramaana encodes the actual statutory text into formal mathematical statements. User queries pass through a proof engine that evaluates the model’s claim against the encoded rules. The system returns a machine-checkable proof if the claim is correct. If the claim violates a rule, the engine identifies the specific breach.

The architecture includes a strict refusal guarantee. The engine blocks answers it cannot mathematically prove. The company states this design prevents the generation of confidently wrong outputs. Technical precedents for this approach include France’s CATALA project for tax formalization and Google DeepMind’s AlphaProof. Teams evaluating AI output in regulated environments typically rely on secondary consensus models, but formal verification offers an absolute mathematical threshold.

Regulated Industry Focus

Pramaana is directing its initial verification layer at four specific verticals where error costs prohibit the use of standard probabilistic models. The company translates complex domain rules into verifiable representations.

Vertical	Application Focus	Key Advisors & Collaborators
Taxation	Verifying US tax code compliance	Danny Werfel, Yale Law School, Stanford
Healthcare	Drug discovery and clinical rule adherence	IIT Delhi, UC Berkeley, Stanford Centaur Lab
Legal	Translating statutes into formal logic	Domain specialists in regulatory compliance
Cybersecurity	Adherence to strict security protocols	Financial and network compliance experts

Developing a deterministic validation layer for these sectors requires extensive domain expertise to translate human legislation into the Lean language accurately.

Technical Leadership

A team of IIT Madras alumni founded Pramaana in 2025. CEO Ranjan Rajagopalan previously launched a verification framework for local search at Google and served as CTO of Astra. Co-founders Krishnan Raghavan and Sanjay Ganapathy Subramaniam serve as CTO and Chief Scientist respectively.

The company is backed by prominent figures in the formal verification space, including Pushmeet Kohli from Google DeepMind and Sriram Rajamani from Microsoft CoreAI. In May 2026, the team published research detailing a TLA+ specified lifecycle for verifying claims within multi-agent coordination patterns. They recently hosted the inaugural Verification Summit in San Francisco, featuring technical panels with representatives from NVIDIA, Microsoft, and DeepMind.

If you build AI systems for regulated industries, evaluate whether your application requires mathematical certainty rather than probabilistic consensus. Encoding domain rules into formal proof languages like Lean provides a defensible audit trail that standard retrieval pipelines cannot match.