Hemispheric

**Product and the concrete problem it solves.** Hemispheric is building what it calls a foundation model for the human brain — an AI system that turns short, non-invasive EEG recordings into objective, quantitative measures of brain state. The concrete problem it attacks is that psychiatry and much of neurology remain diagnosed largely by subjective interview and symptom checklist: PTSD, depression, anxiety, mild traumatic brain injury (mTBI), schizophrenia, and early Alzheimer's are inferred from what a patient reports and what a clinician observes, not from a direct, reproducible physiological readout. That subjectivity produces misdiagnosis, trial-and-error medication cycles, and no reliable way to monitor whether a treatment is working. Hemispheric's answer is a roughly 15-minute test using a dry-electrode EEG headset paired with concurrent behavioral tasks on a tablet, from which its model, Descartes, produces structured biomarkers a physician or researcher can act on. The pitch is not a new sensor but a new interpretive layer — the "brain-decoding" software that sits on top of commodity EEG hardware and extracts signal that traditional clinical EEG reading cannot.

**Core technology and how it actually works.** Descartes is described as a six-billion-parameter, multimodal NeuroAI foundation model trained on more than 250,000 hours of EEG and behavioral data collected from more than 100,000 participants through a global research network — one of the largest privately held brain-activity datasets in the field, per the company. The architecture mirrors the large-language-model playbook: pretrain a large model on a massive, diverse corpus of neural signals so it learns general representations of brain activity, then adapt those representations to specific clinical constructs (a PTSD signature, an mTBI signature, a treatment-response trajectory). Critically, Hemispheric's differentiation is the data asset and the model, not the electrodes: by fusing neural signal with time-locked behavioral responses, the system aims to generalize across people and sites rather than overfit a single lab. The company's stated business model reinforces this — it licenses the proprietary model rather than selling raw data, positioning Descartes as an interpretive engine other diagnostic and pharmaceutical applications can build on.

**Market, customers, and go-to-market.** The initial go-to-market is precision brain health: diagnostic assistance, treatment selection, and progress monitoring across PTSD, mTBI, depression, anxiety, schizophrenia, and Alzheimer's. Hemispheric is pursuing three buyer channels in parallel — clinicians and health systems that want objective psychiatric and neurological measurement; pharmaceutical companies that need objective endpoints and patient stratification for central-nervous-system drug trials (a chronic pain point in CNS drug development, where subjective endpoints sink trials); and, per the company, government bodies. It is targeting U.S. and European markets and has presented to leadership at the FDA's Center for Devices and Radiological Health, signaling a medical-device regulatory path (likely a software-as-a-medical-device clearance strategy) rather than a pure research-tool play. The model-licensing approach is capital-efficient at scale but front-loads the burden of clinical validation and regulatory clearance before meaningful revenue can be recognized.

**Traction, funding, and third-party validation.** Hemispheric emerged from stealth on 15 July 2026 having raised $52 million across three rounds (pre-Seed, Seed, and Series A) over roughly six years of quiet development, and now employs about 112 people spanning neuroscience, AI, software and electronics engineering, and medical systems, with operations across Israel, the United States, and Asia. The investor syndicate is notably deep and brand-name for a company at this stage: Hanaco Ventures, Arkin Capital, OurCrowd, Artofin VC, OneMind/Awareness Capital, Protocol Labs, and L Catterton, alongside prominent individual backers including First Round Capital co-founder Howard Morgan, Naomi Azrieli, Yasmin Lukatz, and Adobe's Scott Belsky. The dataset scale (250,000+ hours; 100,000+ participants), the FDA engagement, and stated work with pharmaceutical partners are the strongest third-party-adjacent validation points, though as of mid-2026 there are no disclosed regulatory clearances, named health-system customers, or peer-reviewed clinical-endpoint results in the public record — the validation is promising but still largely forward-looking.

**Founders and team background.** The founding pair is unusually well-matched to a "foundation model for the brain" thesis. CEO Dr. Hagai Lalazar is a computational neuroscientist, supplying the domain depth in neural signal and brain modeling. CTO Gidi Littwin is a serial AI entrepreneur who co-founded RealFace, the Israeli facial-recognition company acquired by Apple in 2017, and is credited as a co-inventor of Face ID; his background also includes work associated with Apple Vision Pro. That pairing — frontier applied-AI productization plus computational neuroscience — is the credibility core, and the ~112-person team reportedly includes engineers with FDA-cleared medical-imaging and consumer-hardware pedigrees. The principal team question for diligence is clinical and regulatory execution depth: translating a large model into a cleared, reimbursed diagnostic requires clinical-trial, regulatory-affairs, and health-economics muscle distinct from the AI and consumer-tech strengths the founders are best known for.

**Competitive dynamics.** Hemispheric sits at the intersection of two competitive fields. (1) In clinical neuro-diagnostics it competes with quantitative-EEG and brain-network-analytics players (Israel's own ElMindA/BNA lineage, and broader qEEG interpretation), with MEG/fMRI incumbents, and with the entrenched status quo of subjective psychiatric assessment it seeks to displace. (2) In non-invasive neurotech and brain interfaces it shares conceptual space with Kernel and Openwater (non-invasive neuroimaging), Emotiv and Neurable (EEG headsets plus analytics), and Israel's InnerEye (EEG-plus-AI, but oriented to rapid image/threat classification rather than clinical diagnostics). Hemispheric's differentiators are: (i) a genuine foundation-model approach rather than task-specific classifiers; (ii) a large, proprietary, multi-site EEG-plus-behavioral dataset that is hard to replicate and improves with scale; and (iii) a hardware-agnostic, model-licensing posture that lets it ride commodity dry-electrode headsets. The countervailing risk is that a foundation-model narrative is now common, and clinical credibility — not model size — will decide the winner.

**Defense, security, and resilience dual-use relevance.** The dual-use connection here is real but should be stated as adjacency, not a fielded capability. The signature "invisible injuries" of modern combat are precisely two of Hemispheric's first indications: post-traumatic stress disorder and mild traumatic brain injury (blast-related concussion). An objective, portable, 15-minute EEG-based readout for PTSD and mTBI is directly relevant to military medicine, veteran care (e.g., VA-style systems), and allied ministries of defense that struggle to triage, diagnose, and monitor these conditions at scale — and the company's own disclosure that it is working with "government bodies" is consistent with, though not confirmation of, such interest. Beyond clinical care, non-invasive brain-state decoding is adjacent to neuroergonomics: warfighter cognitive-load, fatigue, and fitness-for-duty monitoring, operator readiness, and longer-horizon human-machine-teaming and brain-computer-interface research. The calibrated read: the physiology and the indications make defense/resilience relevance credible and non-forced (human resilience is a strategic capability), but as of mid-2026 there is no disclosed defense contract, military deployment, or dual-use product line — the strategic value is optionality on a clinical platform, not a fielded defense system.

**Growth stage, trajectory, and key diligence risks.** Hemispheric is best read as an unusually well-resourced early-stage deep-tech company: six years of stealth development, a completed Series A within a $52M cumulative raise, ~112 employees, a shipped frontier model, and a large proprietary data moat — but no disclosed regulatory clearance, named commercial customers, or published clinical-endpoint validation. The bull case is a defensible data-and-model asset in a large, underserved brain-health and CNS-pharma market, a credible founding team, and real dual-use optionality via PTSD/mTBI and neuroergonomics. The bear case should dominate diligence: (1) regulatory risk — an AI diagnostic for psychiatric/neurological indications faces a long, uncertain FDA (and EU) path; (2) clinical-validation risk — foundation-model biomarkers must prove prospective, reproducible clinical utility, not just training-set performance, and "world's first" framing invites scrutiny; (3) commercialization and reimbursement risk — clinician adoption and payer coverage for a novel diagnostic are slow and unproven; (4) competitive and scientific risk in a crowded neurotech field; and (5) dual-use is adjacency, so any defense thesis is speculative until a contract or program materializes. Progression toward "mid" would require a regulatory clearance or breakthrough designation, prospective clinical validation, named health-system or pharma customers with disclosed economics, and — for the strategic thesis specifically — a concrete government or defense engagement.