AI models are applied to every dataset under the sun, but they’re inconsistent in their results. That’s as true in the medical world as it is anywhere else, but a startup called Pyramidal thinks it has something certain about a basic model for analyzing brain scan data.
Co-founders Dimitris Sakellariou and Chris Pahuja note that EEG technology, despite being used in nearly every hospital, is fragmented across many types of machines and requires specialized knowledge to interpret. Software that can consistently identify worrisome patterns, regardless of time, location or type of equipment, could improve outcomes for people with brain disorders, while taking some of the burden off overburdened nurses and doctors.
“In the neuro-ICU, there are nurses who monitor the patient and look for signs on the EEG. But sometimes they have to leave the room, and these are acute cases,” Pahuja said. An abnormal reading or alarm could mean a seizure, a stroke, or something else — nurses don’t have that training, and even specialist doctors might recognize one but not the other.
The two started the company after years of working on the feasibility of computational tools in neuroscience. They found that there was certainly a way to automate the analysis of EEG data that would benefit care, but there was no simple way to deploy the technology where it was needed.
“I have experience with this, I mean I’ve been sitting next to neurologists in the operating room understanding why these brain waves are useful, and how we can build computational systems to identify them,” Sakellariou said. “They’re useful in many contexts, but every time you use an EEG, you have to rebuild the entire system for that specific problem. You need to get new data, and you need humans to interpret the data from scratch.”
This would be difficult enough if all EEG systems, hospital IT setups, and data formats were the same, but they varied widely in the most basic elements, such as the number of electrodes on the device and where they were placed.
Pyramidal’s founders believe—and claim to know, though the final outcome of their work has not yet been published—that the basic model of EEG readings could make detecting life-saving brainwave patterns work in an innovative way rather than following months of studies.
To be clear, this is not a comprehensive medical platform—a closer analogue might be Meta’s Llama series of (relatively) open models, which bear the upfront costs of building a basic language-understanding capability. Whether you want to build a customer-service chatbot or a digital friend is up to you, but neither works without a basic ability to understand human language.
But AI models aren’t limited to language—they can be trained to work on fluid dynamics, music, chemistry, and more. For Pyramidal, “language” is brain activity, as read by EEGs, and the resulting model will theoretically be able to understand and interpret signals from any setup, any number of electrodes or type of machine, and any patient.
No one has yet built such a building – at least not publicly.
While they are careful not to overstate their progress so far, Sakellario and Pahuja say: “We’ve built the basic model, we’ve tested it, and now we’re in the process of turning the codebase into a production model that’s ready to scale to billions of parameters. It’s not about research—from day one it’s been about building the model.”
The first production version of the model will be deployed in hospitals early next year, Pahuja said. “We’re working on four pilots starting in the first quarter; all of them will be tested in the ICU, and all of them want to collaborate with us on development.” That will be valuable proof that the model works in the diverse conditions that any intensive care unit faces. (Of course, Pyramidal’s technology will be on top of any monitoring that patients are normally offered.)
The base model still needs to be fine-tuned for specific applications, which Pahuja said they will do themselves at first; unlike many other AI companies, they don’t plan to build a base model and then charge for API usage. But they were clear that it’s still incredibly valuable as is.
“There is no world where a model trained from scratch can perform better than a pre-trained model like ours; starting warm can only make things better,” Sakellariou said. “This is still the largest EEG model ever built, and it’s infinitely larger than anything else out there.”
To move forward, Piramidal needs two things that every AI company needs: money and data. The company is already working on the former, with a $6 million seed funding round co-led by Adverb Ventures and Lionheart Ventures, with participation from Y Combinator and angel investors. The money will go toward computing costs (which are huge for training models) and hiring employees.
In terms of data, they have enough to train their first production model. “It turns out there’s a lot of open source data — but a lot of open source data secluded Data. So we started the process of collecting and curating it into a large integrated data warehouse.
But partnerships with hospitals can provide valuable, large-scale training data—thousands of hours of it. These and other sources could help push the next version of the model beyond human capabilities.
“Right now, we can confidently address this set of specific patterns that doctors are looking for,” says Sakellariou. “But a larger model would allow us to pick out patterns smaller than those that the human eye can consistently and empirically identify.”
This is still a long way off, but breakthrough capabilities are not a prerequisite for improving the quality of care. Pilot ICU projects should allow the technology to be more rigorously evaluated and documented, both in the scientific literature and, more likely, in investor conference rooms.