There is a man who spent ten years sitting on plastic chairs outside hospital clinics in India. He has a PhD from IIT Delhi. He once designed analog circuits in Santa Clara. He is now an Assistant Professor at IIT Mandi. And for most of the last decade, his job was to wait, five hours at a stretch, for a neurologist to finish with patients so he could ask, again, for the EEG recordings the hospital was about to delete the next morning.

That man is Dr. Siddharth Panwar. The pile of “scrap” he collected, one printout at a time, is now the largest proprietary neural dataset of its kind on Earth: over 60,000 hours of continuous EEG from more than 25,000 patients. It is the training fuel for MANAS-1, a 400-million-parameter foundation model that NeuroDx unveiled at the India AI Impact Summit in February 2026, India’s first sovereign foundation model for the human brain.

Check out the video of the conversation here or read on for insights.

The Instrument the West Gave Up On

Electroencephalography is the oldest brain-reading technology that exists, invented by Hans Berger decades before MRI or CT. In the 1960s and 70s, skilled technicians could read tumours and lesions from the wavy lines alone. Then MRI arrived. Structural pictures held up in court. EEG, which requires interpretation, did not. The skill died with its practitioners.

The West, in our opinion, has given up on EEG far too soon.

Siddharth’s frustration is not nostalgic, it is economic. An EEG machine in India costs about ₹3 lakh. An MRI requires tearing down walls to install. EEG is the only brain diagnostic that can realistically reach a primary health centre in rural India. His conceptual leap was to stop treating EEG as a medical chart and start hearing it as language.

I don’t see EEG as anything other than a language. It’s a sequence, very similar to speech. And in fact, there are 20 electrodes, so it’s like a chorus. Twenty people singing.

The implication is large. The same transformer architectures powering ChatGPT, the math of sequential probability and self-supervised reconstruction, work on brainwaves because both are sequences with grammar. MANAS-1 was trained the same way large language models are trained: hide portions of the signal, ask the model to reconstruct them, and let it learn the inherent syntax of neural activity. On the standard benchmark of 25,000 publicly available EEGs that any researcher can scrape, MANAS-1 is the best-performing model in the world. For epilepsy detection, it hits 80 to 90 percent accuracy on the EEG alone, and crosses 95 percent when paired with basic clinical questions.

The Raddi Moat and the Milestones

The competitive advantage is not the architecture. It is the data, and the data is the product of patience.

They delete it every month. It’s just getting washed out of their hard disks. It’s like raddi. Like papers we used to sell. You can buy it raddi ke bhaav.

There were no procurement contracts. Just a decade of waiting rooms and relationships. Eventually, two men noticed. Rajesh Kamra, an IIT Kanpur alumnus and founder of Koovs, and Sandeep Singh, an applied AI veteran from Walmart Labs, brought in their company Newron on a sweat-equity arrangement. Compute, engineers, an office. No venture dilution. Singh is now NeuroDx’s CTO.

🏆 In February 2026, MANAS-1 was unveiled at the India AI Impact Summit. NeuroDx (operating legally as Intellihealth) was selected as 1 of 12 IndiaAI Mission Champions, alongside Sarvam and the IIT Bombay-led BharatGen consortium, and the only physiological foundation model on the list. The model was released open-source on Hugging Face. The full cohort received a two-hour audience with the Prime Minister at his residence.

Taking on Neuralink, with Less Money

Elon Musk’s Neuralink drills holes in skulls. Siddharth listens from the outside. Invasive systems win on signal quality but lose on everything else: surgery, scarring, cost, geography. The global brain-computer interface market is currently valued in the $2.6 to $4 billion range and is forecast to grow at 14 to 18 percent annually through the next decade, with non-invasive systems holding the dominant share. NeuroDx is hardware-agnostic by design, the intelligence layer that can sit on any commercial EEG headset.

He’s doing it invasively. We are trying to do it non-invasively. What he’s doing invasively, we can get 70 to 80 percent of that done without surgery. And we can make it available to a person sitting in a village.

NeuroDx is raising $10 million. In Silicon Valley, this is a rounding error. Meera Murati raised at a billion-plus pre-revenue. Sarvam raised roughly $40 million pre-revenue. NeuroDx is being told $10 million is not possible because they have no earnings yet. Indian deeptech funding grew 58 percent in 2025 to $1.22 billion, and deeptech now represents 15 percent of Indian PE/VC activity, up from 4 percent a decade ago. Yet the appetite for pre-revenue, frontier scientific risk remains thin.

I am willing to die, but I will not scale down my ambition. If my failure has to inspire the next round of great companies, I am fine with that.

The Brain Scan Ahead

The pipeline runs well beyond epilepsy: ADHD and autism for children, depression and anxiety for working-age adults, dementia screening for over-55s. The vision is a five-minute headset session in any clinic that returns a same-day cognitive report. No prescription. No specialist queue. The cost structure of a blood test, not an MRI. MANAS-2, scaling to 2 billion parameters and over 100,000 EEG datasets, is what the round is meant to fund.

The most important AI lab of the next decade may not be in San Francisco or Hangzhou. It may be in a shared office in Bengaluru, run by a man who once sat outside a doctor’s room waiting to be trusted with the trash.

Listen now!

Other ways to listen:

Apple Podcast | Amazon Music

Your Feedback matters

As always, I’d love to hear your thoughts! Whether it’s about this episode or ideas you’ve been playing around with, shoot me an email at ad@thepodium.in. Your feedback keeps these conversations going, and I’m always up for chatting about your startup ideas too.

Until next time,

Your Host,

Akshay Datt

The Founder Thesis Podcast | Learn from disruptive founders

Dr. Siddharth Panwar and NeuroDx: The…