ZoyeMed 3.0,
answered.

Zoyel (ZoyeMed 3.0) is an edge-native autonomous clinical terminal with a longitudinal multimodal AI and an integrated point-of-care laboratory. In a roughly 5-square-metre footprint, it delivers a complete primary-care medical episode - triage, examination, 120+ point-of-care diagnostics, decision support, and prescription support - typically in under 30 minutes. Its primary AI inference runs on the device itself, so core clinical function does not depend on a cloud connection. The kiosk is an integrated technology terminal used together with certified medical devices; the combined system is undergoing Class IIa medical-device certification, and the Zoyel.health software is separately in process for Software-as-a-Medical-Device (SaMD) classification. Final diagnostic and prescribing decisions always remain with a licensed clinician.

It describes how the platform is built: not as a single "AI doctor" model, but as a layered operating system across four functional layers - the Sensorium (acquisition), the Amygdala (safety gate), the Cortex (clinical inference, i.e. the Longitudinal Multimodal Model), and the Thalamus (governance and routing). Each layer runs at a different tempo and answers a different question, and together they provide system-level properties - safety, governance, and longitudinal awareness - that no single model provides on its own.

It is none of the consumer-kiosk or software-only categories. Zoya Technologies positions ZoyeMed as an "Edge AI Clinic": an integrated technology terminal used with certified medical devices - with 120+ direct sensor modalities, on-device AI inference, longitudinal patient-trajectory modelling, an integrated point-of-care lab, and B2B institutional deployment at the scope of a full primary-care encounter (the combined system is undergoing Class IIa medical-device certification). The company's own analysis of thirteen comparable platforms concludes that this specific combination is not occupied by any other currently deployed platform, and invites technical comparison from anyone who believes otherwise. This comparison is intended to define the product category rather than to assert real-time market standing - the landscape keeps moving and not every solution is publicly known.

No. It is a human-in-the-loop clinical decision-support system designed to extend the reach of clinical teams, not replace them. It operates under a trained clinical operator with remote specialist supervision, and autonomous irreversible clinical actions - such as autonomous prescribing or discharge - are prohibited at the architectural level. All consequential decisions are subject to clinician sign-off.

The LMM is the platform's clinical inference layer (the "Cortex"). It works in three stages: (A) multimodal fusion, where each input - vitals, text, ECG and audio waveforms, images, lab values - is encoded and fused into a single per-encounter representation; (B) longitudinal sequencing, where the ordered history of encounters is processed into a patient "trajectory state" (implemented with both GRU-family and State Space Model sequence cores in production); and (C) calibrated multi-task inference, producing a ranked differential set, a triage/urgency classification, risk flags, and recommended next actions, each carrying an explicit confidence score.

A standard LLM or symptom checker treats a patient as a static "snapshot." Zoyel's LMM treats health as a "movie": it tracks the trajectory of vitals, labs, and signals across repeat encounters over time, treating velocity and acceleration of change as first-class clinical signals. The distinguishing property is the "L" (longitudinal) more than the "M" (multimodal) - multimodal fusion is common, but trajectory modelling across years of encounters is the architectural differentiator.

Its core principle is "ground truth before inference." The system acquires objective, device-measured physiological data under its own controls before attempting any inference, rather than relying on patient self-report, which clinical literature has long shown to be an unreliable diagnostic basis. Local multilingual language models are used to converse with the patient, but the clinical reasoning is grounded in measured sensor data.

When the acquired data is insufficient or low-quality, Zoyel withholds a clinical output rather than producing a confident-sounding recommendation on inadequate data. The Amygdala safety gate - part of the data-validation and governance pipelines now deployed in production - scores the sufficiency and quality of inputs and, when they fall short, suppresses inference, returns a reason code, and requests re-measurement. Refusing to answer on bad data is treated as a safety feature.

Through several layers: the Amygdala gate blocks inference on insufficient data; outputs are linked to verified, versioned clinical references and suppressed or labelled when a verified citation cannot be attached; outputs are calibrated and carry confidence scores; and explainability artefacts show which inputs and time windows drove a given recommendation, so clinicians can review the basis rather than trust an opaque output.

The Sensorium spans 120+ clinically validated measurement modalities across six groups: vitals and physical examination (height, weight, BMI, temperature, SpO₂, heart rate, 12-lead ECG, fetal Doppler, digital stethoscopy, dermatoscope, oto-rhinoscope, 4K PTZ camera, thermal imaging, spirometry); biochemistry (liver, renal, lipid, electrolyte, cardiac-enzyme and diabetes panels); immunofluorescence (cardiac markers, inflammation, thyroid, fertility/hormones, tumour markers, vitamins); hematology (3-part and 5-part differentials); AI-vision-analysed rapid diagnostics (e.g. malaria, dengue, typhoid, HIV, Hep-B, Hep-C); and 10-parameter urinalysis. The rapid-diagnostic set is extensible to additional regulator-approved kits on short training cycles.

The confirmed on-terminal examination and imaging modalities include the 12-lead ECG, digital stethoscopy (heart, lung, and gastrointestinal sounds), dermatoscope, oto-rhinoscope, thermal imaging, fetal Doppler, spirometry, and a high-definition multi-camera examination array - alongside the biochemistry, immunofluorescence, hematology, rapid-diagnostic, and urinalysis modules. Capabilities beyond the current published configuration should not be assumed; any future addition would be confirmed by Zoya Technologies and reflected in updated documentation.

The primary AI core is an AMD Strix Halo accelerator with 128 GB of unified memory, dedicated to on-device neural-network inference and to the locally installed language models used for patient conversation. The terminal includes Wi-Fi 7, Bluetooth 5.4, and gigabit Ethernet, with connection bonding and load balancing across multiple links for resilience, plus an internal UPS for safe operation on unstable grids.

ZoyeMed 3.0 runs on a hybrid edge-cloud architecture: privacy-sensitive computation and the primary clinical inference happen locally on the device, so core diagnostic and decision-support functions continue with limited or intermittent connectivity. When a connection is available, only essential de-identified metadata is synchronised; clinical operation does not depend on it.

The Resilience Edition is a distinct, fully standalone configuration engineered to operate with no internet connection at any stage. The complete LMM and the Zoyel.one clinical workflow suite are resident on the device, and every clinical function - triage, examination, point-of-care testing, decision support, and prescription generation - executes locally on the Amygdala edge-AI engine. A terminal can be run indefinitely fully offline; this is a separate product configuration, not a software update to standard units, and it is already in production.

Disaster response and recovery, humanitarian and remote-area delivery, regions with grid instability or network disruption, rural and last-mile settings, maritime and offshore operations, and field medical facilities - settings where connectivity cannot be assumed at the moment clinical capacity is most needed.

Through edge-side anonymisation: patient-identifying data (PHI) is encrypted at rest and stripped on-device before any derived representation is considered for transmission. Raw patient data does not leave the facility unless an explicit authorisation condition permits it; off-device services receive only PHI-clean exports. Zoya Technologies defines sovereignty by this client-side rule - raw PHI stays local - rather than by the location of any server.

Yes. The platform is built to comply with GDPR, HIPAA, and UAE PDPL, alongside local medical-device-software requirements as applicable. Data residency is jurisdiction-pinned, and per-region policy bundles let the system enforce different regulatory, consent, and residency rules per deployment without changing the clinical layer.

Encryption with AES-256 at rest and TLS 1.3 in transit; hardware-protected key management (TPM/HSM-backed); role-based access control (RBAC) and multi-factor authentication (MFA); application sandboxing; and biometric patient-identity verification. Every clinical action, system event, data access, and operator override is recorded.

The immutable audit trail is deployed across the ZoyeMed platform. Every AI inference is logged with its input references, model version, confidence score, and the clinician's disposition (approved, modified, or rejected with a reason), plus a cryptographic hash linking to the previous entry - forming an append-only, tamper-evident hash chain. Default retention is configurable (typically seven years), and required deletions are performed via tombstone events with cryptographic key erasure rather than silent removal. The same audit discipline operates even in disconnected environments, where a cloud audit service is unavailable.

Clinical outputs are packaged as FHIR R4 resources (e.g. Condition, Observation, ServiceRequest, Provenance), with HL7 v2 for legacy integration and DICOM for imaging. Terminology is mapped to SNOMED CT, LOINC, ICD-10 and ICD-11, and RxNorm for medication normalisation. The platform supports facility/jurisdiction profiles such as US Core and the International Patient Summary (IPS), enabling integration with hospital information systems, national health management information systems, and insurance platforms.

It supports three operational modes: doctor-run (a high-tech terminal for physicians), nurse-managed (assisted by on-site staff with on-site or remote clinician sign-off), and autonomous self-service screening. This lets a site operate the terminal with whatever clinical staffing it can support. The interface is multilingual across eight languages.

Agentic actions are bounded by a risk-tier model (informational, draft, propose, critical). Lower tiers can display trends or draft artefacts; higher-risk actions require clinician approval (and may require MFA). Autonomous irreversible clinical actions are prohibited at the architectural level, and when the safety gate blocks inference, automation is limited to requesting re-measurement or routing an escalation.

The Amygdala enforces physiological threat detection. For example, a combination such as low systolic blood pressure with elevated heart rate triggers an immediate clinician alert and typically blocks routine AI output in favour of human attention, with the event written to the audit log.

The ZoyeMed kiosk is an integrated technology terminal that is used together with certified medical devices. Shipped as an integrated setup, it is undergoing certification and is intended to be classified as a Class IIa medical device; the Zoyel.health software is separately in process for Software-as-a-Medical-Device (SaMD) classification. The terminal is CE (Electrical) and NYCE (Mexico) certified, with further clearances in process: CE-MDR at system level, CDSCO (India), COFEPRIS (Mexico), and INVIMA (Colombia). System behaviour and AI output scope are configurable to align with country-specific regulatory, licensure, and scope-of-practice requirements.

Zoyel's AI is the current-generation successor of a 15-year clinical lineage: Glocal Healthcare in India from 2010; the LitmusMX / LitmusDX / Hellolyf clinic-in-a-box platform; and more than 8 million patient episodes across Indian public-health deployments. The predecessor CDSS, LitmusDX, was one of the 13 systems documented in the 2015 peer-reviewed Clinical Decision Support Systems review by Shahsavarani & Abadi. Zoya Technologies is explicit that this is evidence of lineage recognition, not current-generation validation: prospective, peer-reviewed clinical-accuracy studies of the current Zoyel LMM are an ongoing programme of work, not yet complete.

The argument is architectural rather than a published accuracy figure. By grounding decisions in objective, device-measured data ("ground truth before inference"), tracking patients longitudinally rather than as one-off snapshots, refusing to infer on insufficient data, and linking outputs to verified references with calibrated confidence, the system is designed to support better-grounded clinical decisions than self-report-only or single-encounter approaches. Quantified diagnostic-concordance claims await the prospective studies noted above.

The current platform entered commercial deployment in 2025. 44 units have been shipped to Mexico under the MasSalud Clinics programme (with Grupo Salinas as in-country distributor), with units operational and the deployment expanding. Shipments to Colombia are complete with INVIMA approval in process, and contracts are in active negotiation for India, Kazakhstan, and additional markets - all as of 2025/2026, per data available to the company.

More than 23,000 patient consultations have been recorded to date (as of 2026, per data available to the company). A Post-Market Clinical Experience Report covering March 2025–February 2026 documents 18,224 of these under nurse-assisted workflow with remote specialist supervision. Reported results include high operational uptime, strong user satisfaction, zero serious adverse device effects, and zero device incidents involving patient harm; notably, the client-attributable complaint rate (operator training and local infrastructure) exceeded the device-attributable rate, indicating the device itself performed within specification. As a reference workload, terminals across the 15-year lineage have averaged on the order of 40 consultations per day; the current generation is newer and may ramp more gradually.

The full Post-Market Clinical Experience Report (reference ZoyeMed3.0-PMCF-2026-01) is available to qualified regulators and researchers on request, and the detailed Zoyel LMM technical disclosure is available to qualified technical, regulatory, and clinical reviewers under a non-disclosure agreement.

Zoya Technologies holds trademarks for Zoyel and ZoyeMed in India, China, and the UAE, and a registered design patent for the ZoyeMed kiosk in China. A Certificate of Anteriority (via copyright.eu) records the 2025 and 2026 architectural generations - in many jurisdictions copyright arises by self-declaration and is not a regulatory filing, with proof of anteriority being what matters in the event of a counter-claim. A PCT patent application covering the Longitudinal Multimodal Model architecture (including the GRU-family and deployed State Space Model sequence cores, sufficiency gating, and federated-learning governance), together with a patent-of-addition, has been filed and is pending.

It is sold B2B to healthcare networks, government and ministry-of-health programmes, humanitarian organisations, and enterprise operators - through authorised distribution partners. It is positioned for healthcare-access-gap settings in regions such as Latin America, MENA, and Southeast Asia, where primary-care access is the binding constraint, as an infrastructure extension of clinical teams rather than a consumer replacement.

The architecture separates the clinical reasoning stack from a deployment stack that scales across three tiers: edge nodes (the terminals), regional hubs (interoperability gateways, model serving, regional audit and analytics), and a national control plane (governance, model registry, de-identified population analytics, and optional privacy-preserving federated learning under governance controls).

Forward's CarePod was a consumer-positioned autonomous wellness kiosk for US malls, gyms, and workplaces, marketed as an "AI doctor's office." It raised roughly $650 million, reached a $1 billion valuation, launched fewer than ten pods against a 3,200-pod target, and ceased operations in November 2024. The lesson, as Zoya Technologies reads it, was about positioning rather than feasibility: an autonomous system pitched as a replacement for clinical presence in a market that already has working alternatives is a different proposition from the same technology pitched as an extension of clinical presence into settings that have no alternatives. ZoyeMed is an integrated clinical technology terminal - used with certified medical devices, with Class IIa certification of the combined system in process - deployed B2B into healthcare-access-gap settings, operated by a trained operator under specialist supervision - and it carries a 15-year operational lineage rather than a three-year consumer pivot.

Babylon reached a peak valuation above $4 billion, reported a $369.8 million operating loss on $1.1 billion of revenue in 2022, and filed for Chapter 7 in August 2023; its Rwanda deployment reached near-national scale before ceasing. The architectural lesson is specific: a scaled cloud-native platform with no proprietary ground-truth-acquisition layer depends entirely on patient self-report, which caps clinical accuracy below what the scale demanded. ZoyeMed's answer is its proprietary sensorium - "ground truth before inference," acquiring objective device-measured data rather than relying on what a patient describes over a video call. The Edge AI Clinic category is, in part, the architectural response to that lesson.

ZoyeMed is deliberately not in the "unbounded-autonomy" category that Zoya Technologies identifies as the one that has repeatedly failed. It is human-in-the-loop: a trained operator with remote specialist supervision; an Amygdala gate that refuses to produce output on insufficient data; a prohibition - enforced at the architectural level - on autonomous irreversible actions such as prescribing or discharge; an immutable audit trail; and final decisions resting with a licensed clinician in every mode.

Those are cloud-native, scaled telemedicine and tech-enabled-primary-care platforms with little or no direct sensing, where AI augments workflow and clinical depth comes from human clinician networks. Zoya Technologies notes that scaled telemedicine without proprietary device architecture is margin-constrained - several such platforms continued to post losses even at very large covered-life counts. ZoyeMed operates in a different layer of the stack: a clinical device that acquires its own data and runs inference on-device. The relationship is complementary rather than competitive - a ZoyeMed-class terminal could feed clinical-grade data into a platform of Ping An's class.

They are the closest architectural peers and have executed excellently within their scope; the distinction is scope, not quality or ambition. TytoCare is a home-based, roughly six-modality guided-examination toolkit; Butterfly is a handheld single-modality (ultrasound) device. ZoyeMed is a full clinical terminal - 120+ modalities including an integrated point-of-care laboratory - operating in professional settings at the scope of a complete primary-care encounter, with longitudinal trajectory modelling across encounters. One framing: the depth of edge AI those devices achieve within a single modality, extended across 120+ integrated modalities and a full encounter.

This is acknowledged openly. Ada's 50+ independent peer-reviewed publications represent a depth of external validation that Zoya Technologies explicitly says ZoyeMed aspires to and is investing in. ZoyeMed's current-generation independent validation is earlier in its arc: the predecessor CDSS (LitmusDX) was one of the 13 systems documented in the 2015 peer-reviewed CDSS review, and prospective, peer-reviewed clinical-accuracy studies of the current Zoyel LMM are an ongoing programme of work that is not yet complete. The company's position is that acknowledging this honestly strengthens rather than weakens the claims it does make.

The company has been built from operating income, with no external equity raised to date. Zoya Technologies frames this as economic discipline derived from operational maturity rather than venture funding, and contrasts it with the consumer-autonomy thesis that consumed billions (Forward, Babylon) without producing sustainable returns. It makes no representation about future financing. Large-scale deployment outside Mexico is acknowledged as the next thing to be demonstrated: Colombia is in progress pending INVIMA clearance, with India, Kazakhstan, and other markets in active contracting.

The company is direct about its limits. It does not claim to have "solved primary care"; it does not claim its deployment numbers exceed peers' (Ping An has served far more patients); it does not claim the 2015 LitmusDX citation validates current performance; it does not claim the 18,224-consultation Post-Market Clinical Experience Report demonstrates clinical accuracy (it demonstrates safety and operational feasibility); and it avoids "world's first" claims. Open questions it names include large-scale deployment outside Mexico, prospective clinical-accuracy studies, multi-region closed-loop outcome integration, federated learning at multi-country scale, and long-horizon model-drift analysis - all treated as future work.