A co-creator of Apple’s FaceID and Vision Pro technologies has decided that deep learning, once trained to unlock iPhones via facial recognition, can now crack the "black box" of human cognition. Gidi Littwin, who left Cupertino in 2020, has spent the last six years building a neural network model to decipher the brain's electrical activity. The irony is unmistakable: while traditional medicine has relied for decades on either invasive surgery or subjective questionnaires, the Hemispheric project is attempting to turn mental health into a pure engineering challenge solved by a non-invasive headset.
Industrializing Brain Data
Littwin’s transition from consumer electronics to MedTech is not a random career pivot but a direct technical evolution. At Apple, he oversaw massive data collection operations, involving hundreds of thousands of subjects to refine hand-tracking algorithms. He has brought that same assembly-line approach to Hemispheric, securing the company’s primary asset and competitive moat: a dataset of 250,000 hours of brain activity recordings from 100,000 paid volunteers. In an industry where studies often rely on samples of just a few dozen people, this is more than a database; it is a formidable barrier to entry for a company backed by $52 million in investment. Hemispheric's models derive the brain's functional state from electrical signals with the same clinical, statistical efficiency that LLMs use to derive meaning from text.
"These projects were backed by colossal data collection operations, and we knew that at Hemispheric, we would have to build something similar," Littwin states.
Company co-founder Hagai Lalazar originally sought a way to study the brain without opening the skull, screening around 75 candidates before hiring Littwin for his expertise in commercializing complex technologies. Today, capital is being deployed to move EEG out of specialized clinics and toward frontier health management models.
From Subjective Observations to Scalable Algorithms
The project’s ambition is a radical paradigm shift in treating Alzheimer’s, Parkinson’s, and PTSD. Currently, doctors must rely on external behavioral signs—a method whose accuracy leaves much to be desired. Hemispheric's model is already demonstrating precision in diagnosing schizophrenia and depression within test groups. The workflow is designed to be consumer-friendly: a 15-minute session wearing a lightweight headset while interacting with a tablet app. The AI decodes the signals, helping clinicians tailor therapy and monitor progress, effectively turning elite neurological expertise into scalable software.
The primary bottleneck for this "productized" approach remains regulatory hurdles. Early next year, Hemispheric plans to submit its first product for PTSD treatment for FDA approval, targeting a mass launch in 2027. A narrow specialization built on a massive proprietary dataset points toward a strategy of infrastructure dominance: Littwin isn't building an abstract superintelligence, but a specialized conveyor belt. The goal is provocatively simple—to make deep neurological diagnostics as accessible and affordable as a standard blood test. By applying Apple’s playbook to neurology, Hemispheric is trying to strip the subjectivity out of psychiatry. If Littwin clears the FDA’s bureaucratic gauntlet by 2027, the brain will finally become a readable interface, and AI will pivot from tracking our faces to predicting our cognitive decline.