Google Research: SensorLM
Sensor-language models trained on 59.7M hours of wearable data from more than 103,000 people (NeurIPS 2025). Text descriptions are generated from the sensor data alone, with no speech. Paper.
Focus: describing the signal.
Company
Why SeroState exists
After the 2023 earthquakes, the people carrying everyone else kept saying “I’m fine.” No one could see what their bodies were carrying.
The body is measured at scale.The moment is not.
Wearables now capture physiology for millions of people. The data shows that a heart rate rose, not that it rose when someone said “No, I’m fine.” That moment is where risk hides.
The missing layer: Session Physiology
SeroState brings context-labelled physiology, what we call Session Physiology, to behavioral-health sessions. Heart rate comes first, with voice as its core companion, HRV when sampling allows and skin response (EDA) when the wearable provides it. All of it is aligned to what was said in the session. SeroState flags where words and body disagree, scores each flag for confidence and signal quality, and delivers it as EHR-ready (FHIR R4) data beside the session note, for clinician review. The clinician always decides. Hardware-agnostic. Stateless processing: your organization holds the PHI; we don’t store it.
What others have built
Wearable and AI research and platforms
Apple: Wearable Behavior Model
A foundation model trained on about 2.5B hours of Apple Watch behaviour data (ICML 2025).
Focus: long-term behaviour patterns.
Verily: Pre
An AI and data platform where health data is unified, governed and analysed.
Focus: data infrastructure.
Empatica
A medical-grade wearable platform with FDA-cleared measures, including skin response (EDA), used in 120+ clinical trials.
Focus: clinical-grade signals.
WHOOP
A consumer health platform with 2.5M+ members, valued at $10.1B (2026), adding blood testing to wearable data.
Focus: body plus lab data.
Hume AI
Emotional-expression models for voice, offered to developers.
Focus: expression in voice.
Behavioral health AI
Eleos Health
AI notes and analytics from therapy-session audio, inside the health record, used by 120+ organisations.
Focus: session documentation.
Upheal
An AI-native therapy health-record system with notes from session recordings.
Focus: session documentation.
Lyssn
AI quality review of therapy and crisis-line conversations, with an NIMH grant.
Focus: conversation quality.
Ellipsis Health
Voice biomarkers in an AI care manager for health plans.
Focus: voice.
Canary Speech
Voice biomarkers for health systems.
Focus: voice.
Limbic
An AI assistant for NHS mental-health intake, certified as a Class IIa UKCA medical device.
Focus: text intake.
Figures as published by each source, as of October 2026. SeroState is not affiliated with these companies.
SeroState adds what sits between them: the body and the voice, in the moment, in context.
Where SeroState sits
Data platforms such as Verily Pre are not plotted.
SeroState’s view, based on each company’s published materials (linked above). Illustrative positioning, not a measurement.
What’s already in place
- Method
- Session Physiology definition and science foundation, published. Science · Session Physiology.
- Reference device
- Built and tested against the Corsano CardioWatch, a certified medical wearable. Hardware-agnostic by design.
- European route
- Selected for Erasmus for Young Entrepreneurs 2026, hosted by InPerson Group.
- Founder’s earlier work
- The model was first used under UMAXLIFE, the founder’s previous venture: 16+ licensed therapists in post-earthquake Hatay with Rotary International and CATED, and at Beykent University Dental Hospital. UMAXLIFE won Sanofi Care4Rare #1 (January 2025).
- Product
- API and EHR-ready (FHIR R4) output defined; sample session report available. Sample report →
European route
"Healthcare systems across the DACH region want solutions that are human-centric, data-driven, and scalable. SeroState delivers exactly that. We’re putting our full ecosystem behind bringing it to European providers."
On UMAXLIFE, the founder’s previous venture
“UMAXLIFE is an elegant, streamlined solution to measure quality biomarkers of stress and assign context to the physiological changes. By transforming voice into time-locked events, it lets us link context with stress responses, finally making it possible to create and track actionable patient strategies.”
On UMAXLIFE, the founder’s previous venture
“UMAXLIFE’s innovative analytics have been a game-changer at our hospital. By measuring stress, it empowers patients and clinicians. Integrating mental health into dental care resulted in consistently high patient satisfaction scores.”
What’s next
We’re raising to run the first consented pilot and validation study.