Hemispheric, a brain-diagnostics startup co-founded by Apple FaceID co-inventor Gidi Littwin, has raised $52 million after spending six years training a frontier AI model on 250,000 hours of brain data collected from 100,000 people. The company aims to use the model to diagnose depression, PTSD, Parkinson's, Alzheimer's, and schizophrenia from EEG signals — replacing subjective questionnaires with a 15-minute headset scan. Hemispheric will submit its first product, a PTSD diagnostic, to the FDA in early 2027, with a public rollout targeted for later that year.
The 100,000-subject dataset is the core asset. Volunteers across Asia, Tel Aviv, and Boston were paid to complete game-like tasks that activated different regions of the brain while EEG hardware recorded electrical activity. That data trained a generalized model that infers brain function from electrical signals the way a large language model infers meaning from text.
Littwin left Apple in 2020, where he had helped ship FaceID and was working on hand-tracking for Vision Pro. Both projects required what he described as hundreds of thousands of subjects' worth of training data — an operational template he brought directly to Hemispheric. Co-founder Hagai Lalazar had already begun building AI for non-invasive brain analysis and cold-messaged Littwin on LinkedIn after interviewing roughly 75 candidates for a commercial co-founder.
Key facts
- 01Hemispheric raised $52 million after training a frontier AI model on 250,000 hours of brain data from 100,000 paid volunteers.
- 02Co-founder Gidi Littwin co-invented Apple's FaceID and worked on Vision Pro hand-tracking before leaving Apple in 2020.
- 03The company will submit its first product, a PTSD diagnostic, to the FDA in early 2027, targeting a public rollout later that year.
- 04A patient wears a lightweight EEG headset for about 15 minutes while interacting with a tablet app to generate a diagnosis.
- 05Investors include early Uber backer Howard Morgan alongside American and Israeli venture firms.
The data-collection challenge is the harder half of the business. Diagnosing cognitive disorders has historically relied on clinician-administered questionnaires and behavioral observation because each brain's electrical signature is idiosyncratic. Hemispheric's bet is that a model trained on a large enough dataset can find the pattern beneath the noise.
In testing, the team said the generalized model made accurate deductions about brain health on subsets of subjects diagnosed with PTSD, schizophrenia, and depression. A clinical study is now underway to test whether the same model can diagnose and even predict Alzheimer's before symptoms present.
The commercial pitch is speed and cost. A patient wears a lightweight EEG headset for about 15 minutes while interacting with a tablet app; the model then helps clinicians decode signals, select a likely-effective treatment, and monitor progress.
AI-assisted diagnostics are already in clinical deployment for conditions like lung cancer across Europe, and OpenAI and Anthropic are expanding into healthcare — intensifying competition for smaller specialists. Hemispheric's angle is that standard EEG hardware was never designed for deep learning workloads, so the company is building its own brain scanners to produce cleaner training and inference data.
Investors in the $52 million round include American and Israeli venture firms and individual backers, among them early Uber investor Howard Morgan. The proceeds will fund partnerships with governments, health systems, and pharmaceutical firms; US hiring; the FDA submission; and continued data collection targeting millions more subjects.
“These devices were never built for machine learning and definitely not deep learning.”— Gidi Littwin, Hemispheric co-founder
The risks are the usual ones for AI-in-medicine. FDA clearance for a novel diagnostic device is a multi-year process, and the strength of the training set does not guarantee generalization across populations that were under-represented in it. A model that performs well on the initial cohort could still miss diagnoses in patients whose neurological baseline differs, and the clinical utility of predicting conditions like Alzheimer's years in advance depends on whether effective early interventions actually exist.
If Hemispheric clears the FDA, the strategic implication is that the FaceID playbook — massive proprietary data collection paired with vertically integrated hardware — travels well beyond consumer devices. The frontier-lab race is dominated by general-purpose model builders, but the more valuable niches over the next five years may belong to companies that own a defensible dataset in a regulated domain. A 15-minute EEG that costs less than a specialist visit would reshape the economics of mental-health diagnosis, and it is the kind of vertical AI product the horizontal labs are unlikely to build themselves.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
