Today, I'm thrilled to announce Pramaana's $27M seed, led by @khoslaventures. The foundational domains that hold the world together: tax, law, finance, healthcare; all run on certainty. Probabilistic AI can't give them that. We’ve been asked to accept wrong answers with AI as ‘hallucinations’, while in traditional software terms, it’s just a bug. And a wrong answer in such mission-critical domains is more than just a bug, it's a liability that could have catastrophic impact. We built Pramaana to deliver a 100% trustable experience to the domains that run on certainty: AI that is provably correct, not probabilistically correct. We turn statute and regulation into machine-verifiable code, so every output ships with mathematical proof of correctness. Our mission is to make AI take ownership of it’s work. Pramaana in Sanskrit stands for “means of valid knowledge”, and we’re going to achieve that by formalizing the world’s knowledge.

What if we could do this not just in math but for every knowledge system? That's what we're building @PramaanaLabs

AI is entering math. We’re ensuring math enters everywhere else! Building the models to make this a reality at @PramaanaLabs.

Current Frontier LLM models hallucinate with Tax queries on phase-outs, miss quadratic interactions, and leave money on the table; all while sounding confident. Formal verification of these outputs at run-time is the only meaningful way to address this. Cell-by-cell 1040 modeling machine-checked proofs that your tax plan is compliant and provably optimal. Audit-defensible. Fearless. Authored by: Yoshiki Takashima Read how we turn complex tax optimization into verifiable truth → pramaanalabs.ai/blog/audit-d…

When mathematicians enter other domains, they solve it. The most popular example of that is Jim Simons and Renaissance Technologies "solving" algorithmic trading to a large extent. Now we are entering an arena where mathematicians are looking for a new home as AI is encroaching theirs. Extremely interesting times as I expect multiple hard real world problems to be solved by them over the next few years. When theory stops being exciting, great practical things would be achieved. We at @PramaanaLabs are one such group of mathematicians, betting that the next wave of hard, real-world problems gets solved by people who think in proofs. We are actively speaking with more mathematicians, and the hunger and energy are unmistakable. If you are one, reach out. Let us talk about what we could build together. @krishnan_rag @sanjaygsub

The way AI markets itself with such confidence convinces people it can be blindly trusted. In reality, these are just probabilistic machines with zero guarantees of correctness. Nothing ensures that any given sentence or reasoning is actually true. To deliver on AI abundance, we must stop guessing and use formal models to mathematically prove correctness.

It's the tax season of the year for India. We formalized India’s ITR-2 tax logic in Lean 4 and made it a tool for all to use, with a mathematical proof trail to aide your choice of regime selection. Feel free to tinker with our tool before you file your returns: taxoptimizer.pramaanalabs.ai… Created by: @manoj_sure @techieShukla

How do we make robot policies robust to rare but high-impact failures? Video #World #Models (WMs) are rapidly becoming a powerful tool for robotics, enabling policy evaluation and improvement by "imagining" future outcomes. But there's a catch: these imagined futures are typically nominal samples, making it easy to overlook the rare yet safety-critical events that matter most. In our new paper, StressDream: Steering Video World Models for Robust Policy Evaluation and Improvement, we explore a simple but powerful idea: 💡 Instead of passively sampling futures, actively steer world model imaginations toward high-impact yet still plausible scenarios. StressDream optimizes the initial diffusion noise at inference time, allowing us to generate targeted stress-test scenarios without retraining the world model. This enables: - More robust policy evaluation by exposing failure modes that random sampling often misses. - Improved policy optimization by training against challenging but realistic imagined futures. As generative world models become a foundation for #Physical #AI, the ability to systematically probe their "long tail" of plausible futures will be increasingly important for building reliable and trustworthy autonomous systems. 📌 𝖯𝗋𝗈𝗃𝖾𝖼𝗍 𝖯𝖺𝗀𝖾: junwon.me/StressDream/ 📄 𝖯𝖺𝗉𝖾𝗋: arxiv.org/abs/2606.00267 Work led by Junwon Seo, with a great set of collaborators: Sushant Veer, Thomas Ran Tian, Wenhao Ding, Apoorva Sharma, Karen Leung, Edward Schmerling, Andrea Bajcsy. @NVIDIADRIVE @NVIDIAAI #Robotics #WorldModels #PhysicalAISafety #AISafety #AutonomousSystems #RobotLearnin

Is the output audit-defensible is the only question you should be asking your AI models. Audit-defensibility lies at the very core of what we're building at Pramaana for Tax.

Was an interesting conversation yesterday ! Thanks for hosting me at the Startup Street segment - @CNBCTV18Live youtu.be/Awh5ZNSRIYQ?si=SVX5…

Hallucinations in Medical AI can very well be the difference between life and death ! Verified AI in healthcare and medicine a much-needed unlock !

Here's our take on specialized provers paving the way ahead for verification in AI

Vinod Khosla on the two most important things to get right when building a company The first thing for a startup to get right is make sure they’re inventing something radically different: “If you’re extrapolating the past to predict the future, the big companies do that very well. Intel knows how to go from a 15nm process to 10 to 7 to 5… Entrepreneurs fundamentally have to invent the future they want in a very different way, and you have to have conviction.” For this reason, Khosla Ventures tends not to invest in startups that are trying to do something “slightly better” (e.g. slightly lower cost, slightly more efficient, etc.) because the risk is too high. As Vinod explains: “If you’re fundamentally reinventing something and changing the assumptions, then you have a lot more time to get it right so there’s a lot more iterations you can do.” Which brings Vinod back to the second thing every founder has to get right: build the right team. “If you get the right team — it doesn’t even matter if you get the spec right — it will iterate to the right answer, and almost all good startups do that. This is why the team is more important than the market you pick.” However, you must make sure to foster a culture of debate and disagreeing: “If there’s a single founder and you’re dictating terms, you’re not going to evolve… You have to set up a culture so that it doesn’t matter if you’re senior or junior — you speak up if you have a concern or an idea. You’ll detect problems much earlier, and as a group you can evolve your direction.” Vinod points out that this is one of the reasons Andy Grove started the casual, dress-down culture of Silicon Valley in the 70s: “He wanted an engineer — whether they were junior or senior or a VP — to speak up and not be intimidated that the executives were all wearing ties and coats.” Source: @southpkcommons (Apr 2025)

The most complex phenomena arise from scalable recombination of very simple rules. Whether it's galaxies, chips, or neural networks, if you find the right primitive building blocks, the complexity takes care of itself.

AI in 2040 will not be built on the stack we are using today. It will be much closer to optimal. The current stack has 3-4 orders of magnitude of data inefficiency and 4-5 orders of magnitude of compute inefficiency. Near-optimal AI is what symbolic learning will deliver.

Sridhar Vembu received the distinguished alum award at IIT Madras in 2016, the year I graduated. I distinctly remember seeing him for the first time, giving a talk on entrepreneurship to our graduating CS batch. A decade on, to hear him give a shout out to @PramaanaLabs and our thesis on national TV is definitely a goosebumps moment! Thank you @svembu 🙏 Let's bring efficient, verified intelligence to the world!

Thanks for the shoutout @svembu

Thanks for the shout out @svembu let's bring verifiable super intelligence to the world!

A mirror facing a mirror creates the illusion of depth. A prediction machine directed towards itself creates the illusion of direction.

Good marketing is finding different ways to tell the truth.