Ranjan Rajagopalan has raised $27 million for Pramaana Labs, a young AI company trying to put formal verification between large language models and the rule-heavy work where a wrong answer can become a liability.
The seed round, reported Wednesday by TechCrunch, was led by Khosla Ventures, with participation from Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound. Pramaana Labs has not disclosed a valuation, revenue, customers, or the close date of the round. The most concrete public date is the June 17, 2026 announcement.
The founder profile matters here because Pramaana Labs is not pitching AI reliability as an enterprise compliance add-on. Rajagopalan's background runs through systems work. IIT Madras' computer science alumni page says he completed a BTech-MTech dual degree in 2017, worked at Graviton Research Capital and Google, and at Google launched a verification framework to improve local-search freshness. In 2023, he co-founded Astra as CTO, where the same IIT page says he designed and tested an RLVR sales agent for automated CRM updates.
That path explains the company's bet: for some domains, the answer is less important than the evidence chain behind it. Pramaana Labs says it is building domain-specific systems that translate tax codes, clinical guidelines, legal rules, safety constraints, policies, scientific text, and regulatory language into machine-checkable logic. Its website describes a stack that formalizes domain knowledge, searches the solution space with provers and solvers, and returns proof artifacts that experts can inspect.
The product is a verifier, not a chatbot
Pramaana Labs' first market is tax, according to both TechCrunch and the company's own site. The company lists taxability, nexus, exemptions, credits, statutory interpretation, and policy simulation among its initial use cases. TechCrunch reported that Pramaana Labs is working with former IRS commissioner Danny Werfel on tax law, while professors from IIT Delhi, IIT Madras, and UC Berkeley oversee cybersecurity and drug-discovery systems.
The logic is straightforward. Tax law is structured, exception-heavy, and expensive to get wrong. It is also a domain where the rules already exist, but in prose written for human experts rather than for machines. Rajagopalan told TechCrunch that once a codified version exists, reasoning on top of it can become deterministic.
That is the core distinction Pramaana Labs is trying to sell. A conventional LLM can still generate the natural-language answer, but Pramaana Labs wants a deterministic layer above or around it that checks whether the answer follows from formalized rules. The company says its Domain Formalizer converts regulatory, statutory, policy, and scientific text into precise representations capturing definitions, conditions, exceptions, dependencies, and source traceability.
The pitch is not that models will stop being probabilistic. It is that in high-stakes workflows, the probabilistic part should not be the only artifact a lawyer, tax adviser, clinician, scientist, auditor, or regulator has to evaluate.
A founding team built around verification and applied AI
Pramaana Labs' founding bench is unusually aligned with the problem. The Verification Summit site, hosted by Pramaana Labs in San Francisco on June 10, 2026, lists Rajagopalan as co-founder and CEO. It lists Krishnan Raghavan as co-founder and CTO, and Sanjay Ganapathy Subramaniam as co-founder and chief scientist.
Raghavan's public Glean bio says he joined Glean in April 2022 as an engineering leader focused on machine intelligence, structured data representations, reasoning, and verification. It also says he previously worked at Google on the freshness and accuracy of Google Maps and holds a computer science and engineering degree from IIT Madras. The Verification Summit page says he built Glean's India search team.
Ganapathy's summit bio says he was previously a staff research engineer at Google DeepMind, where he architected Gemini's tool-use system and led post-training work. The company is essentially assembling people who have worked on search freshness, enterprise AI reliability, tool use, and formal systems, then pointing them at sectors where an LLM answer without auditability is not enough.
There is also an India-US company footprint around the team. TheCompanyCheck lists RKS Pramaana Labs India Private Limited as incorporated on September 12, 2025 in Bangalore South, Karnataka, with Rajagopalan and Raghavan as directors. The public records do not by themselves establish the full legal relationship to the San Francisco company, but they show the founders were formalizing an India entity months before the funding announcement.
Why Khosla is in the round
Khosla Ventures' role is not incidental. Pramaana Labs hosted its Verification Summit one week before the TechCrunch story, with Vinod Khosla listed as anchoring the event and Khosla partner Kanu Gulati moderating a panel. The summit framed verification as a technical field moving from outputs that sound right to outputs that can be proved right.
The need is not theoretical. The RuleArena benchmark, published as an ACL 2025 paper, tested LLMs on rule-guided reasoning across airline baggage rules, NBA transactions, and tax regulations. Its authors found that models struggled to identify and apply the right rules, made computation errors even after identifying relevant rules, and generally performed poorly on the benchmark.
Pramaana Labs is positioning itself against that failure mode. The company's site argues that confidence scores and citations do not provide accountability when the task requires showing what rule was used, what assumptions were made, what constraints were checked, where the answer follows, and where it fails.
The hard part is formalizing the messy world
The risk in Pramaana Labs' model is not whether formal verification works in principle. It does. The risk is whether formalizing tax, law, drug discovery, clinical safety, cybersecurity, and policy domains can be done quickly enough, accurately enough, and cheaply enough to support a venture-scale company.
TechCrunch noted France's CATALA project, which formalizes parts of tax and benefits law into executable code, as precedent Rajagopalan points to. That precedent cuts both ways. It shows the approach is plausible. It also shows why this work is laborious: statutes and regulations are full of jurisdictional definitions, exceptions, dependencies, edge cases, and policy judgments.
Pramaana Labs' answer is to make formalization itself an AI-assisted workflow. If its Domain Formalizer can reduce the cost of translating real-world rules into checkable systems, Pramaana Labs could sit underneath the next generation of AI products in regulated industries. If expert review remains the bottleneck, the company may end up as a high-end services business wrapped around a technical platform.
The $27 million seed gives Rajagopalan and his co-founders room to find out. It also fits a broader funding pattern: investors are still willing to write large early checks for AI infrastructure companies that promise reliability, not just generation. RuntimeWire reported in May that AI search upstarts were pulling major capital as founders tried to rebuild retrieval around LLMs. Pramaana Labs is making the adjacent bet that retrieval is not enough in domains where rules decide the outcome.
The open question is commercial proof. Pramaana Labs has disclosed the investors and the technical thesis. It has not disclosed paying customers, pilots, pricing, ARR, or deployment timelines. For a seed-stage lab, that is not unusual. For a company promising provable AI in industries where mistakes are expensive, those details will matter quickly.