Physical AI Clinical Infrastructure · Class IIa Regulated

The clinic that
sees the patient.

Physical AI clinical infrastructure that acquires objective physiological ground truth before inference.

Built on 15 years of clinical deployment across more than 9 million patient episodes.

15Years Clinical AI Lineage
9M+Patient Episodes
120+Sensor Modalities
18,224PMCF Consultations
0Cloud Required for Inference

The Structural Problem

Healthcare AI has no senses.

Current AI systems depend on what the patient says. But patients are often unreliable historians - filtered by perception, vocabulary, stoicism, and the limits of self-observation. Clinical inference requires objective physiological data, not narrative.

LLMs are blind and deaf.

A patient describing "gastric pain" may be experiencing gastritis. Or an inferior-wall myocardial infarction. Without a 12-lead ECG and troponin measurement, no AI - regardless of how capable - can reliably distinguish the two.

The reasoning capacity of modern AI has outpaced the quality of the data it reasons on. The bottleneck is not intelligence. It is sensory grounding.

Same Patient · Two Views
Patient Says
"I've had some chest discomfort. Probably acidity."
Body Shows
ST elevation, lead II/III/aVF
Patient Says
"I feel fine. Just a check-up."
Body Shows
BP 168/104, HR 96, SpO₂ 94%
Patient Says
"Sugar's under control. I take my meds."
Body Shows
HbA1c 9.2% · ↑ from 7.8% (90d)
Deficit 01

The Chatbot Fallacy

Text-only AI reasons from words alone. "Patient Speak" without sensor data is not clinical input. It is noise dressed as signal - and decisions built on it are structurally unsafe.

Deficit 02

Snapshot Medicine

A single lab value misses trajectory. The velocity of biomarker change is often more informative than the value itself. Medicine sees snapshots. It should see trajectories.

Deficit 03

The Clinical Memory Problem

Doctors carry patient context in their heads. When a doctor is unavailable, the context is lost. A new clinician sees only the file. The system must hold the trajectory.

A New Category

From chatbot AI to Physical AI.

A clinical AI system must do more than reason. It must sense, validate, reason longitudinally, govern, and escalate to a human clinician. This requires more than software. It requires infrastructure.

Existing Approach

Chatbot AI

  • Hears only what the patient says
  • Depends on cloud inference
  • Treats each encounter as a snapshot
  • Has no objective clinical acquisition layer
  • Cannot refuse to answer when input is inadequate
  • No governance layer between AI and clinical action

ZoyeMed Approach

Physical AI

  • Acquires data directly from the body - 120+ modalities
  • Reasons on-device - no cloud dependency for primary inference
  • Tracks patient trajectory across the full health history
  • Validates data quality before inference is permitted
  • Refuses to produce output when inputs are insufficient
  • Governance and audit layer enforces clinician sign-off

Operational Evidence

Fifteen years in the field. Not fifteen months in a lab.

Before the architecture matters, the operational record matters. ZoyeMed is the current instantiation of a clinical AI lineage that has been in continuous public-health deployment since 2010.

Operational Lineage

  • 15Years continuous clinical AI deployment India 2010 → present
  • 9M+Patient episodes across the platform lineage
  • 1,100+Hospital beds operated WB · UP · Bihar · Odisha

Deployment Evidence

  • 44Units delivered, Mexico institutional programme
  • 18,224PMCF-documented consultations Mar 2025 – Feb 2026
  • 3Active country deployments Mexico · Colombia · Malaysia

Regulatory Progress

  • CE Electrical certified
  • NYCE Mexico certified
  • CE-MDR system-level in process
  • CDSCO India in process
  • COFEPRIS Mexico · INVIMA Colombia in process

Recognition

  • WEF / Schwab Foundation Social Entrepreneur of the Year 2020
  • UN Innovation Public Appreciation Award 2020
  • Bloomberg New Economy Forum Gamechanger 2020
  • Frost & Sullivan India Telemedicine 2020

Research & IP

  • CDSS lineage cited in peer-reviewed literature 2015
  • Longitudinal multimodal model architecture PCT patent filed 2025 - 26
  • Clinical white paper published Feb 2026
  • Technical disclosure available under NDA

System Architecture

Four layers. One clinical operating system.

ZoyeMed is structured as a layered architecture where each component has a distinct function and runs at a different tempo. No single model. No black box. Each layer can refuse, escalate, or block.

L1 · SENSORIUM
Physical acquisition layer
120+ clinically validated sensor modalities - biochemistry, hematology, immunofluorescence, 12-lead ECG, digital stethoscopy, thermal imaging, spirometry, dermatoscope, urinalysis, AI-vision rapid diagnostics. Time-aligned to sub-50ms precision within encounter windows.
L2 · AMYGDALA
Safety & data-quality gate
Scores data sufficiency across modalities before any inference. Refuses output when inputs are inadequate. Blocks unsafe inference. Anonymises patient identifiers on-device. The architecture of refusal.
L3 · CORTEX
Longitudinal multimodal model
Per-modality encoders fuse into a visit embedding. A proprietary sequence engine processes visits across the patient's full history - tracking biomarker velocity, not just point values. Calibrated ICD-10/11 differentials, urgency, risk flags, next-action recommendations.
L4 · THALAMUS
Governance & audit layer
Per-jurisdiction policy bundles enforce data residency, consent scope, protocol routing, and operator credentialing. Immutable cryptographic audit trail. Clinician sign-off required for all consequential actions. Autonomous prescribing architecturally prohibited.
L5 · INTELLIGENCE
Population intelligence layer
Anonymised real-time signal layer derived from aggregated terminal data. Disease incidence, drug resistance, nutritional deficiency, outbreak signals - all from clinically measured ground truth, not survey reports.
Why a longitudinal model

A single biomarker value can mislead. Trajectories reveal risk. A proprietary sequence architecture tracks the velocity and acceleration of biomarker change across a patient's full history - detecting adverse trends invisible in single-encounter data. Most preventable adverse clinical events have precursor trends of exactly this kind.

Why edge inference

Primary AI inference runs on-device. Patient identifying data is anonymised at the edge - identifiers never reach the inference layer. Cloud round-trip is not required for clinical operation. This is what makes data sovereignty a property of architecture, not policy.

Clinician authority preserved. ZoyeMed provides structured clinical decision support. Final clinical decisions and prescriptions remain under licensed clinician authority. Autonomous irreversible clinical actions are architecturally prohibited at the system level.

Ground Truth Before Inference

A patient says "acidity." The system measures.

The architecture in motion - through a single clinical scenario. Patient self-report enters the system. Sensor data overrides it. The safety gate escalates. A clinician decides.

Step 01 · Patient Narrative

"I have some acidity. Burning chest. Probably just gas."

Patient presentation begins with self-report. A chatbot would reason from this text alone. ZoyeMed treats it as input - not as ground truth.

Step 02 · Sensorium Acquisition

The body is measured.

12-Lead ECGcaptured
Troponin-I0.84 ng/mL
BP92/58
HR118 bpm
SpO₂91%
Thermal · L armflagged
Step 03 · Safety Gate

The narrative is overridden.

ECG · ST elevationII, III, aVF
Troponinelevated
Patterninferior wall MI
AcuityEMERGENT

"Acidity" is recoded. AI output suppressed pending clinician review.

Step 04 · Clinician Action

The clinician decides.

Remote specialist alerted. Local operator briefed. Emergency referral protocol triggered. Patient routed to nearest interventional cardiology centre. The system does not act autonomously. It surfaces, escalates, and waits for the human.

Longitudinal Intelligence

Medicine sees snapshots. Zoyel sees the movie.

A single lab value can look reassuring. The trajectory may be dangerous. The system tracks not just where the patient is - but where they are heading, and how fast.

HbA1c, 5.8%. A single measurement. Within the "normal" range. A snapshot model concludes: no action required.

Three months earlier: 5.0%. The velocity is the signal. The patient is moving toward dysglycaemia faster than the threshold is approaching.

A longitudinal model detects this. A snapshot model cannot. Most preventable adverse clinical events have precursor trends of exactly this kind.

Similar pattern: creatinine drift within normal range. SpO₂ baseline shift. Thyroid trajectory across pregnancy. Statin response over 12 weeks. The trajectory carries the diagnosis.

HbA1c · Single patient · 3 visits

Snapshot view vs trajectory view 7.0% 5.7% 5.0% 4.0% Visit 1 Visit 2 Visit 3 normal range 5.0% 5.4% 5.8% trajectory flagged
Snapshot view:normal range - no action
Trajectory view:+0.27% / 90d - early intervention

National Health Intelligence

One terminal is a clinic. A national deployment is a health intelligence layer.

A single ZoyeMed terminal produces individual patient data. A national fleet produces a real-time epidemiological signal - derived from clinically measured ground truth, not survey self-report.

  • - Outbreak detection
  • - Chronic disease mapping
  • - Anaemia & nutrition surveillance
  • - Occupational health clusters
  • - Drug resistance trends
  • - Payer risk analytics
  • - Regional disease burden
  • - Resource allocation
Discuss National Deployment →

Live Health Intelligence Stream

HbA1c velocity trending ↑ - District 7 cohort
Dengue RDT positivity +2.3% - Northern region, 30d
SpO₂ baseline shift - Mining district, flagged for review
Troponin-I elevation cluster - Occupational cohort
Paediatric anaemia - Rural 14.2% vs urban 9.8%
Statin prescription gap - 38% eligible, 9% prescribed
ICD-10 A15 trajectory cluster - Cross-district TB pattern
HPV vaccination coverage drop - Region 4, age cohort 9–14

Illustrative. All patient data anonymised at edge before aggregation. Data residency configurable per jurisdiction.

15-Year Clinical Lineage

No venture capital buys fifteen years.

The architectural choices in ZoyeMed 3.0 are answers to questions asked repeatedly in places where getting it wrong is measured in patients - not quarterly metrics. Nabarangpur. The Sundarbans. Rajasthan. The questions were hard. The answers took time.

2010

Indian Entity

Dr. Azim - Founder of the Indian entity - establishes a model to deliver standardised, process-engineered primary care across rural eastern India. First site: Sonamukhi, Bankura district, West Bengal.

2015

LitmusDX - Clinical Decision Support System

Bayesian semantic CDSS developed and deployed across the Glocal network. Cited in peer-reviewed Clinical Decision Support Systems literature as among the more visible CDSS in global operation at the time. Network: 1,100+ hospital beds across four states.

Peer-reviewed CDSS citation 2015
2016

Sundarbans - First Hybrid Platform Deployment

The Sundarbans: tidal rivers, island-scattered villages, populations a day's boat-journey from a physician. The architecture learns to function where connectivity is intermittent, power is unreliable, and patient self-report is structurally unreliable.

2017–19

Rajasthan · Odisha · Madhya Pradesh - National Scale

~100 clinics under India's National Health Mission, Rajasthan. Odisha Digital Dispensary Project in Nabarangpur - one of India's most underdeveloped districts. State MoU for 102 dispensaries across 23 districts. ~500 dispensaries in Madhya Pradesh.

World Rural Healthcare Conference Award 2018 Express Public Health Award 2018
2020

HellolyfCX · Pandemic Terminal · International Recognition

Pandemic-capable autonomous terminal with integrated disinfection deployed across Indian states. Five simultaneous international recognitions in a single year - independent validation of operational and clinical credibility.

WEF / Schwab Social Entrepreneur UN Innovation Award Bloomberg New Economy Gamechanger Frost & Sullivan
2022–24

Zoya Technologies, Dubai - Architecture Reengineered

Complete re-engineering for edge-first inference, longitudinal multimodal AI, clinical decision support guardrails, FHIR R4 output. PCT patent filed for the longitudinal multimodal model architecture. First 44 commercial units delivered to Mexico.

2025–26

ZoyeMed 3.0 - Commercial Deployment at Scale

22,000+ consultations through end-March 2026. 18,224 documented in a published Post-Market Clinical Experience Report. Signed multi-year deployment agreements in Colombia. Malaysia entity and IP registration with MDEC underway.

WHX Dubai 2026 BusinessWire coverage

Market Position

Four categories solve part of the problem. ZoyeMed integrates the parts.

The digital health market has sorted into four operating categories. Each addresses one dimension of clinical care delivery. ZoyeMed is a fifth - a physical AI clinical infrastructure layer that integrates sensing, edge inference, longitudinal intelligence, governance, and institutional deployment.

Category 1

Telemedicine Platforms

Extends physician reach over video and chat at consumer or enterprise scale. AI as workflow augmentation. Cloud-native, no direct sensing.

  • Reach without acquisition
  • Workflow AI, not deep CDS

Category 2

Software-Only Clinical AI

Validated clinical decision support delivered as regulated software, sitting inside existing clinical infrastructure. No proprietary sensing layer.

  • Depth without sensors
  • Depends on hospital data quality

Category 3

Narrow Regulated Devices

Class II / FDA-cleared devices with deep edge AI within a single clinical function - handheld ultrasound, guided home examination, continuous monitoring.

  • Sensors without scope
  • Excellent within bounded scope

Category 4

Screening & Access Infrastructure

Physical access points in healthcare-gap settings. Modest modality counts. Deliberate scope discipline.

  • Access without depth
  • Live clinicians, not autonomous CDS

Category 5 · ZoyeMed

Physical AI Clinical Infrastructure

Integrates sensing, on-device inference, longitudinal trajectory modelling, governance, and institutional deployment into one clinical operating system.

  • Class IIa regulated medical device
  • 120+ direct sensor modalities incl. full POC lab
  • On-device longitudinal AI inference
  • 15-year CDSS lineage · 9M+ patient episodes
  • B2B institutional deployment only

A separate sub-category - unbounded consumer autonomous kiosks attempting wide clinical scope without regulation, sensing depth, or CDSS lineage - has produced multiple high-profile closures over 2023–24. The full landscape analysis, including a thirteen-platform seven-axis comparison, is available in the Technical White Paper under NDA.

Deployment Model

Built for institutions. Not consumers.

ZoyeMed deploys as a fleet. Typical deployments begin at 100+ units. Per-client configuration handles all jurisdiction-specific compliance requirements at the governance layer - without branching the codebase.

Sovereign healthcare infrastructure.

For Ministries of Health and national public-health agencies. Extend clinical access into underserved geographies while building a real-time epidemiological intelligence layer over the deployment.

Discuss National Programme →
  • Access expansionClinical-grade encounters in geographies without physicians. Operator-staffed terminals supervised remotely by specialist clinicians.
  • National health intelligenceReal-time epidemiological signal derived from objective sensor data - disease incidence, drug resistance, nutritional deficiency, outbreak detection.
  • Data sovereignty by architecturePHI anonymised at edge. In-country residency for aggregated data. Configurable per-jurisdiction governance bundles.
  • Standardised qualitySame diagnostic depth at every deployment site, regardless of operator background or local infrastructure.

Extend reach. Standardise quality.

For hospital networks and integrated health systems. Distribute diagnostic depth into spoke catchments while preserving central clinical authority and audit.

Discuss Network Deployment →
  • Distributed diagnostic reachFull point-of-care lab and 120+ modality examination at any spoke site. Specialist time leveraged across many more encounters.
  • Standardised data captureFHIR R4 output. ICD-10 / ICD-11 / SNOMED CT / LOINC interoperability with existing HIS, EMR, and PACS systems.
  • Specialist leverageRemote specialists oversee 5–10× more encounters per hour. AI handles the routine; specialists handle the consequential.
  • Auditable clinical governanceImmutable audit trail. Every AI inference, every clinician disposition, every override - logged with cryptographic integrity.

Objective data. Longitudinal trajectory.

For health insurers and payer organisations. Replace self-reported health declarations with objective sensor-derived data. Detect risk earlier. Price members on what is - not what is reported.

Discuss Payer Integration →
  • Objective pre-authorisation dataSensor-acquired ground truth at the point of clinical encounter. Eliminates the gap between self-report and clinical reality.
  • Early risk detectionLongitudinal trajectory model surfaces drift toward chronic conditions before threshold-based diagnosis. Earlier intervention, lower long-term cost.
  • Longitudinal member analyticsPopulation-level risk pattern detection across the insured cohort. Resource allocation informed by objective trajectory data.
  • Fraud reductionVerified, sensor-anchored clinical events. Cannot be retrospectively fabricated.

Workforce health, on-site.

For industrial, occupational health, and corporate networks. Deploy clinical-grade health monitoring at industrial scale - across mines, factories, oilfields, smart cities, and large corporate campuses.

Discuss On-Site Deployment →
  • Occupational risk surveillanceReal-time detection of occupational disease patterns - respiratory function shifts, biomarker cluster anomalies, occupational toxin exposure indicators.
  • Workforce health monitoringPeriodic comprehensive health encounters for every worker. Longitudinal cohort tracking across years of employment.
  • Regulatory complianceAuditable, FHIR-interoperable record of workforce health monitoring. Meets occupational health standards in regulated industries.
  • On-site clinical accessWorker encounters without site evacuation. Trained operator with remote specialist supervision.

Architecture, interoperability, governance.

For technical reviewers, regulatory teams, and integration partners. The full technical disclosure - covering longitudinal multimodal model architecture, safety gate design, PHI handling, and interoperability - is available under NDA.

Request Technical Disclosure →
  • Longitudinal multimodal architecturePer-modality encoders, visit embedding fusion, sequence engine, multi-task calibrated inference. PCT patent filed 2025–26.
  • InteroperabilityFHIR R4 resources. SNOMED CT, LOINC, ICD-10, ICD-11, RxNorm terminology mapping. Native integration with major HIS / EMR platforms.
  • Security & privacyEdge-side PHI anonymisation. Cryptographic audit chain. Per-jurisdiction policy bundles. Configurable data residency.
  • Validation pipelineMulti-stage safety gates. Each stage can independently block output. Reported accuracy is the floor - not the final clinical performance after guardrails.

Active Deployments

Mexico - 44 units delivered under institutional programme; multi-phase contract up to 300 units; 18,224 PMCF-documented consultations through Feb 2026.  |  Colombia - Signed agreement for annual deployments over a three-year term; first ZoyeMed 3.0 unit delivered to Bogotá.  |  Malaysia - Entity and IP registration with MDEC underway; in-country data residency–compliant infrastructure confirmed.

Evidence & Research

For serious reviewers. Go deeper.

Documents are tiered by sensitivity. Public materials are downloadable directly. Technical materials require email-verified institutional access. Commercial and IP materials require an enterprise briefing call.

White Paper · Feb 2026Public

Overcoming the Cognitive & Sensory Deficit in Modern Medicine

Dr. Syed Sabahat Azim, MBBS, Ex-IAS

The clinical rationale for Physical AI. Covers the Snapshot Fallacy, the safety gate, longitudinal multimodal modelling, and evidence-based prescribing democratisation. Grounded in 15 years and 9 million patient outcomes.

↓ Download PDF
Working Monograph · SSRNPublic

Beyond Primary, Secondary, and Tertiary: A 2×2 Healthcare Framework

Dr. Syed Sabahat Azim · WEF 2020

A practice-derived replacement for the century-old Alma-Ata tier classification. Two axes - acuity and intervention dependency - generate four operationally coherent quadrants with distinct planning implications.

Technical White Paper · Apr 2026Verified Access

The Category Nobody Else Is Building

Team Zoyel · Zoya Technologies LLC

Thirteen-platform competitive analysis using a seven-axis framework. Includes Post-Market Clinical Experience Report summary covering 18,224 documented consultations. Available to institutional reviewers with verified corporate email.

3D Terminal Model · InteractiveVerified Access

ZoyeMed 3.0 - Interactive 3D Terminal

Hardware visualisation

Explore the ZoyeMed 3.0 terminal in 3D. Sensor positioning, operator workflow, footprint, integration ports. Available to institutional reviewers with verified corporate email.

PMCF Full Report · 2026NDA + Call Required

Post-Market Clinical Experience Report

Mexico institutional programme, Mar 2025 – Feb 2026

The full PMCF report covering 18,224 consultations. Patient-level safety, operational performance, and AI-clinician concordance data. Released after a discovery call and NDA execution.

Financial Model · 2026NDA + Call Required

Deployment & Financial Model

For institutional buyers and investors

Unit economics, deployment scaling, multi-country revenue model, capital requirements. Released after a qualification call and NDA execution. Not delivered via form.

IP Technical Disclosure · NDANDA + Call Required

Longitudinal Multimodal Model - Patent & IP Disclosure

PCT filed 2025–26

Multimodal fusion, sequence architecture, sufficiency gating, agentic human-in-the-loop orchestration. For qualified technical, regulatory, or clinical reviewers under signed NDA.

About access tiers. Public materials are freely downloadable and citable. Verified-access materials require a corporate-email submission and are delivered automatically via verified email. NDA + Call materials require a discovery or briefing call before disclosure - protecting the integrity of commercial and IP-sensitive information. All requests are logged. We follow up.

Enterprise Engagement

Deploy physical AI clinical infrastructure.

For governments, hospital networks, insurers, and public-health systems. Typical deployments begin at 100+ units. Qualified partners receive full technical disclosure, PMCF report, and financial model under NDA.

ZoyeMed® is a registered trademark of Zoya Technologies LLC. Class IIa Medical Device. CE Electrical · NYCE Mexico certified. CE-MDR, CDSCO, COFEPRIS, INVIMA in process. Longitudinal multimodal model architecture subject to PCT patent filing 2025-26. Technical disclosure available under NDA.