Pramaana Labs is a San Francisco-based AI startup building a verification layer for artificial intelligence , aimed at high-stakes regulated domains where probabilistic outputs are not acceptable. The company translates complex domain knowledge like tax codes, clinical guidelines, legal statutes, and safety constraints into formally verifiable representations , so that an AI answer ships with a machine-checkable proof against the underlying rules. The approach is akin to the LEAN programming language used to verify mathematical proofs , applied to statutes and regulations rather than pure math.
This launch marks the company's emergence from stealth alongside a $27 million seed round led by Khosla Ventures, with participation from Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound . According to the team, the capital will go toward training its formalisation and prover models, hiring more AI research talent, and scaling domain expertise , with initial focus on statutory tax reasoning, legal compliance, healthcare safety, and autonomous systems . The timing tracks with growing pressure on AI deployments in regulated work, including courtroom scrutiny of attorneys who file briefs containing fabricated citations.
The company was co-founded by Ranjan Rajagopalan, Krishnan Raghavan, and Sanjay Ganapathy Subramaniam , with Rajagopalan serving as CEO. As he framed it in the launch announcement, Pramaana in Sanskrit stands for "means of valid knowledge," and the company aims to achieve that by formalising the world's knowledge , turning regulation into code that an AI system can be held accountable to.
Comments (12)
Sign in to join the discussion.
N
Nilufar QodirovaJun 18, 2026
Pramaana means 'proof' in Sanskrit. Whoever named this company actually thought about it for more than 4 seconds, which already puts you ahead of 90% of AI startups.
D
DeShawn O.Jun 18, 2026
Tax, law, finance, healthcare in one sentence is the most ambitious wedge I've seen this quarter. Pick one and I'll believe you, pick four and I'll wait for the case study.
Y
yusufJun 18, 2026
The launch video pacing was actually clean, but the cut to the whiteboard at 0:18 felt like a Khosla deck template. Lose the b-roll, trust the voiceover.
M
Marta K.Jun 18, 2026
Before any of this matters, where does the inference happen and can I keep PHI in a specific region? Asking on behalf of every general counsel about to email you.
V
Vikram I.Jun 18, 2026
Skipping the technical debate. SOC2 Type 2? HIPAA BAA? SSO via Okta and Entra? If yes, my procurement team can talk Tuesday.
K
Kenji L.Jun 18, 2026
'AI that proves its work' is fine. 'AI you can cite in court' is the one you wanted. No charge.
P
Priya R.Jun 18, 2026
Three weeks out from my own evals startup launch and I see this post. I'll be in the corner rethinking my entire roadmap if anyone needs me.
O
obafemiJun 18, 2026
Every 'verifiable AI' demo I've seen has been a lookup table with extra steps. Show me one adversarial input where the proof doesn't collapse into 'trust us, the retriever found it.'
S
Søren HalvorsenJun 18, 2026
Curious if there's an open core here or if this is closed weights and closed verifier. The legal teams I talk to want to inspect the proof layer themselves.
A
Anya BrennanJun 18, 2026
Is the 'proof' here closer to formal verification, neuro-symbolic grounding, or just calibrated retrieval citations? The tweet thread waved at all three and committed to none.
T
Tomasz P.Jun 18, 2026
The market for 'AI in regulated industries' isn't growing, it's bifurcating. Half will buy a horizontal platform, the other half will build in-house and you get the smaller half.
R
ravenJun 18, 2026
The wordmark on pramaanalabs.ai has the two A's kerned like they're avoiding each other at a wedding. Otherwise the palette is restrained in a good way.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Pramaana Labs
AI that proves its work
AI that proves its work
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Pramaana Labs is a San Francisco-based AI startup building a verification layer for artificial intelligence , aimed at high-stakes regulated domains where probabilistic outputs are not acceptable. The company translates complex domain knowledge like tax codes, clinical guidelines, legal statutes, and safety constraints into formally verifiable representations , so that an AI answer ships with a machine-checkable proof against the underlying rules. The approach is akin to the LEAN programming language used to verify mathematical proofs , applied to statutes and regulations rather than pure math. This launch marks the company's emergence from stealth alongside a $27 million seed round led by Khosla Ventures, with participation from Accel, BoldCap, Nexus Venture Partners, Premji Invest, and Unbound . According to the team, the capital will go toward training its formalisation and prover models, hiring more AI research talent, and scaling domain expertise , with initial focus on statutory tax reasoning, legal compliance, healthcare safety, and autonomous systems . The timing tracks with growing pressure on AI deployments in regulated work, including courtroom scrutiny of attorneys who file briefs containing fabricated citations. The company was co-founded by Ranjan Rajagopalan, Krishnan Raghavan, and Sanjay Ganapathy Subramaniam , with Rajagopalan serving as CEO. As he framed it in the launch announcement, Pramaana in Sanskrit stands for "means of valid knowledge," and the company aims to achieve that by formalising the world's knowledge , turning regulation into code that an AI system can be held accountable to.
Comments (12)
Pramaana means 'proof' in Sanskrit. Whoever named this company actually thought about it for more than 4 seconds, which already puts you ahead of 90% of AI startups.
Tax, law, finance, healthcare in one sentence is the most ambitious wedge I've seen this quarter. Pick one and I'll believe you, pick four and I'll wait for the case study.
The launch video pacing was actually clean, but the cut to the whiteboard at 0:18 felt like a Khosla deck template. Lose the b-roll, trust the voiceover.
Before any of this matters, where does the inference happen and can I keep PHI in a specific region? Asking on behalf of every general counsel about to email you.
Skipping the technical debate. SOC2 Type 2? HIPAA BAA? SSO via Okta and Entra? If yes, my procurement team can talk Tuesday.
'AI that proves its work' is fine. 'AI you can cite in court' is the one you wanted. No charge.
Three weeks out from my own evals startup launch and I see this post. I'll be in the corner rethinking my entire roadmap if anyone needs me.
Every 'verifiable AI' demo I've seen has been a lookup table with extra steps. Show me one adversarial input where the proof doesn't collapse into 'trust us, the retriever found it.'
Curious if there's an open core here or if this is closed weights and closed verifier. The legal teams I talk to want to inspect the proof layer themselves.
Is the 'proof' here closer to formal verification, neuro-symbolic grounding, or just calibrated retrieval citations? The tweet thread waved at all three and committed to none.
The market for 'AI in regulated industries' isn't growing, it's bifurcating. Half will buy a horizontal platform, the other half will build in-house and you get the smaller half.
The wordmark on pramaanalabs.ai has the two A's kerned like they're avoiding each other at a wedding. Otherwise the palette is restrained in a good way.