Piramidal

What is a brain foundation model?

We are building a foundation model, from the ground up, trained on a massive and diverse corpus of brainwave data i.e. Electroencephalography (EEG). In other words, similar to how Large Language Models (LLM) are trained on human language ie. text, our model is trained on brain language ie. brainwaves.

Our foundation model will be able to understand and detect the many facets of brain activity, from disease to wellbeing. Crucially, similar to how ChatGPT can be fine-tuned to become a lawyer or a python developer, our brain foundation model will easily be fine-tuned to become a neurologist co-pilot or a biomarker discovery assistant. And, this is only our first milestone - more on this below.

The opportunity

Our brain foundation model will address a variety of use cases with minimal training, such as:

• Diagnosis of a vast majority of neurological conditions.
• Drug and biomarker discovery in clinical trials.
• Remote monitoring of brainwave abnormalities, e.g., treatment response to drugs or sleep irregularities.
• Real-time tracking of brain health, such as cognition or head injuries, measured through helmets in sports and military.
• Empowering devices such as Apple Vision Pro and Meta Quest to translate brain signals into everyday actions.

Why are we starting with neurological conditions?

During his PhD at London's St Thomas' EEG department, Dimitris discovered a key insight: EEG is a crucial test for neurological disorders, but its manual review process is both slow and error-prone. Specifically:

• Neurologists manually inspect the EEG brainwaves to identify abnormal patterns. This can lead to a 30% margin of error and crucially misdiagnoses.
• There is a substantial rise of EEG data being collected, due to the deployment of  remote EEG devices. Despite these remote devices continuously recording patients at home, there is no way to review this enormous and continuous stream of incoming data.
• Conversely to the upward EEG data trend, there is an extreme shortage of neurologists leading to wait times lasting several months.

Our first product

Our first product is a copilot for epilepsy diagnosis. The product, which is powered by our brain foundation model, will assist neurologists by obtaining diagnostic insights around a patient’s EEG, in real time. This approach cuts down the time to review EEG from hours to seconds. Additionally, our standardized EEG analysis reduces the margin of error from manual inspection, which in many cases leads to misdiagnosis and recurrent hospital visits.

Shortly after epilepsy, we will fine tune our foundation model to provide diagnosis for a vast majority of other conditions such as sleep and cognitive disorders.

Who we are

We are Dimitris and Kris. Dimitris has a PhD in neuroscience and AI, and has been working on EEG for 12 years, but also shipped production level AI as a deep learning lead at IQVIA. Kris has a product background, building healthcare and AI products across Google and Spotify. We are both avid musicians and love punk rock.

Our ask..

We are looking for intros to:

1. Neurologists, sleep doctors, neurotech companies, or companies in clinical trials in order to sign LOIs. We can automate the analysis of the brainwave (EEG) data at a fraction of their current cost.
2. Academic or healthcare institutions that store EEG data. In exchange for accessing their EEG records, we will provide our model for free for research purposes.

You can send them our way to info@piramidal.ai!

Please also share our post and spread the world - you never know who it might help! 🖤