Apple's FaceID Architect Launches Brain-Health AI Model Trained on 100,000 EEG Datasets

Gidi Littwin’s Pivot From Computer Vision to Brain Diagnostics

Gidi Littwin, who co-invented Apple’s FaceID biometric system and helped develop hand-tracking technology for the Vision Pro, has transitioned into medical-device development through his startup Hemispheric. According to Wired AI, Littwin departed Apple in 2020 and subsequently co-founded Hemispheric with neuroscientist Hagai Lalazar, who had independently begun developing machine-learning methods to analyze brain electrical activity without surgical intervention. The startup has now secured $52 million in funding to advance its core product: a deep-learning model that infers brain health from non-invasive electroencephalogram (EEG) recordings.

Littwin’s background in large-scale data collection proved directly transferable. At Apple, he oversaw “hundreds of thousands of subjects’ worth of data” collection to train computer-vision systems for facial and hand recognition. Hemispheric replicated this operational approach by assembling a dataset of 250,000 hours of brain-activity recordings from 100,000 paid volunteers distributed across geographic regions including East Asia, Tel Aviv, and Boston.

How Hemispheric’s Brain Model Works

The startup trained its frontier AI model on this EEG corpus to perform a task analogous to language-model inference: statistical pattern-matching across high-dimensional neural signals. Volunteers completed interactive activities designed to activate distinct brain regions, generating training examples for the model to learn the relationship between electrical activity and cognitive states.

According to Wired, Hemispheric tested its generalized model against subsets of subjects with confirmed diagnoses of post-traumatic stress disorder (PTSD), schizophrenia, and major depression, reporting that the model made accurate inferences about brain health in these populations. The team is currently running a clinical trial to evaluate whether the model can diagnose and even predict Alzheimer’s disease progression.

FDA Pathway and Clinical Rollout Timeline

Littwin and Lalazar plan to submit their first product—a PTSD diagnostic tool—to the U.S. Food and Drug Administration in early 2027. According to Wired, public availability is targeted for late 2027. The diagnostic workflow involves a patient wearing a lightweight EEG headset for approximately 15 minutes while interacting with a tablet application; the model then assists clinicians in decoding signals to support diagnosis, predict treatment efficacy, and track therapeutic progress.

Why This Matters

Hemispheric’s approach addresses a longstanding clinical gap: cognitive disorders including depression, Alzheimer’s, and Parkinson’s have historically relied on subjective assessment tools, behavioral observation, and expensive or invasive procedures like positron-emission tomography (PET) imaging. A validated, non-invasive EEG-based diagnostic model could democratize access to objective biomarkers, particularly in underserved regions lacking advanced neuroimaging infrastructure.

The success of Littwin’s model on independent test populations is a critical threshold for clinical adoption—if results hold across diverse patient cohorts and external validation studies, FDA clearance could establish a new standard-of-care pathway for cognitive-disorder screening. The timeline to late 2027 suggests the startup expects preliminary clinical-trial results to support the regulatory submission, though independent reproduction of the model’s generalization performance will be essential before widespread clinical deployment.