The co-inventor of Apple Face ID And VisionPro technology has spent the last six years building a frontier artificial intelligence model that could one day help decode the brain’s electrical activity to diagnose cognitive disorders.
Today, Gidi Littwin’s startup Hemispheric raised $52 million after collecting brain data from 100,000 people to train deep learning models to examine the brain without resorting to invasive procedures.
Littwin left Apple in 2020, looking for a change. He found it when his Hemispheric co-founder Hagai Lalazar sent him a cold message on LinkedIn. Lalazar had begun developing artificial intelligence to study the brain without resorting to surgery and was looking for a commercially minded co-founder to take the company forward. By the time he found Littwin, he had spoken to about 75 candidates.
Littwin had helped develop FaceID and was at the time working on hand tracking for an augmented reality product, the Vision Pro. As part of this, he had to collect what he told WIRED were “hundreds of thousands of data points from subjects” to train the deep learning models that power the technology.
“Behind these projects were massive data collection operations and we knew we had to build something very similar at Hemispheric,” Littwin says, “and we did.”
Because each individual’s brain activity is different, doctors have had to rely heavily on subjective questionnaires and behavioral observations to diagnose depression, Alzheimer’s disease, and Parkinson’s disease. To get around this problem, Littwin and Hagai collected their “most valuable asset”: a quarter of a million hours of brain data from 100,000 paid volunteers across Asia, as well as in Tel Aviv and Boston. The subjects undertook a series of activities that resembled games but activated different parts of their brain.
This data helped train a frontier model, which infers brain function from electrical activity in the skull in the same way that large language models infer meaning by statistically analyzing text. They then tested the generalized model on subsets of people, including those diagnosed with PTSD, schizophrenia and depression, and said the model made accurate inferences about individuals’ brain health. The team is currently working on a clinical study to test whether their model can diagnose and even predict Alzheimer’s disease.
The team will submit its first product, which will be used to study PTSD, to the FDA for approval early next year. They hope this will allow them to roll out the product to the public later in 2027.
To help diagnose a cognitive disorder, a patient wears a lightweight EEG headset that measures electrical activity in the brain for about 15 minutes while interacting with an app on a tablet. Hemispheric says its AI model will then help clinicians decode signals to make diagnoses, select the most effective intervention by making treatment predictions, and monitor progress.
“The future we envision is one where this is akin to a blood test,” Lalazar says. “The device will be very, very inexpensive; it can be sold and distributed to mental health clinics, hospitals and even psychologists’ offices.”
AI-assisted diagnostic tools for diseases such as lung cancer are already in clinical use and are accelerating access to treatment across Europe. Meanwhile, AI giants including OpenAI and Anthropic are expanding into the healthcare sector, intensifying competition to attract many startups to the field.
Hemispheric raised seed funding from investors including US and Israeli venture capital firms and individual investors, including Howard Morgan, Uber’s first backer. They will use the money to advance partnerships with governments, health organizations and pharmaceutical companies, hire more in the United States and work to obtain regulatory approval. They also plan to measure more brain data from millions of people in an effort to improve their model.
The two are also developing their own brain scanners to obtain information that the company says can provide more useful data for their models than traditional EEGs. “These devices were never designed for machine learning and certainly not for deep learning,” says Littwin.