VEHA Health Care Education LLP seal VEHAHealth Care Education LLP

AI platform · Product 05 · In development

Clinical Intelligence

AI drafts. Doctors decide. A clinical operations platform where AI does the writing-up and the specialist reach extends past the building — while the medical decision, and the accountability for it, stays exactly where it belongs.

In active development. Not yet available for clinical use.

1
Continuous clinical record, bedside to specialist
FHIR R4
Primary integration contract
100%
AI output labelled
0
Clinical decisions made without a clinician

In development. This page describes a platform being built, not one you can buy today. It is not available for clinical use, and the compliance frameworks below are design commitments rather than completed certifications.

01 Primary workflow

From nurse intake to
doctor-approved note.

The documentation burden moves off the clinician and becomes a review step — without the record ever leaving clinical control.

  • 01

    Patient context

    Staff register or select a patient, assign unit and bed, and open the relevant clinical episode or ICU context.

  • 02

    Nurse intake

    Vitals, symptoms, assessments, attachments and bedside notes are captured on the mobile app, close to the patient and in structured form.

  • 03

    AI assistance

    The platform organises that clinical data and prepares draft summaries, documentation support, abnormal-trend highlights and review prompts.

  • 04

    Doctor review

    A doctor reviews the record, edits the AI-drafted content, approves or rejects it, and finalises the note or care-plan update.

  • 05

    Audit and handoff

    Every critical action is logged append-only, routed to the right queue, visible to authorised teams and available for reporting.

02 AI governance

The AI writes.
It does not decide.

An AI that quietly edits a medical record is a liability, not a product. So the constraint is built into the data model rather than into a policy document: AI content is labelled, editable and traceable, and it cannot enter official documentation without a named clinician approving it.

A final note stores its author, its approver, the timestamps, the source data behind it and every edit along the way. Escalation suggestions stay advisory until an authorised clinician acts.

Governance rule

AI produces draft support only. The medical decision, the approval and the accountability remain with an authorised clinician.

Audit rule

Capture the prompt context, the generated draft reference, the clinician edits, the final approval, any rejection reason and the downstream handoff status.

Safety rule

Disclaimers, review states, permission checks, visible sources and escalation guardrails all sit in front of production use.

03 Remote care model

Specialist depth,
without losing ownership.

Most Indian hospitals cannot staff an intensivist around the clock. Tele-ICU extends that reach — but the local team keeps clinical ownership; the remote specialist advises, and the platform tracks whether that advice was accepted, modified or declined.

Case escalation

Staff flag a patient for specialist review on severity, abnormal vitals, unit protocol or manual escalation.

Remote specialist review

Specialists get structured patient context, recent trends, draft summaries, attachments and the consult question in one case view.

Recommendation capture

Recommendations are documented, routed back to the local doctor and marked accepted, modified, declined or pending clarification.

Operational oversight

Leaders monitor pending consults, response times, bottlenecks and handoff completeness.

04 Capabilities

What the platform does

AI-assisted documentation

AI drafts clinical summaries, progress notes, abnormal-trend highlights and handoff summaries from structured bedside data — turning the documentation burden into a review task rather than a writing task.

Doctor approval built in

Every draft lands in an approval queue. A doctor can edit, accept, reject, request clarification, or mark an output as not clinically useful. Nothing reaches the record unapproved.

Tele-ICU and specialist review

Bedside teams escalate a case; remote specialists see structured context, recent trends and the consult question in one view, and their recommendation is routed back and tracked to closure.

Traceable by design

Each final note stores its author, approver, timestamps, the source data behind it, the clinician's edits and the full audit trail. AI content is visibly labelled and never silently merged in.

Speaks the hospital's language

HL7 FHIR R4 as the primary contract, with HL7 v2 and DICOMweb where that is what a site exposes — normalised at the boundary so core services see one model.

Tenant isolation throughout

Every patient record, workflow, session, audit event and report carries hospital scope. RBAC drives what a user sees, and the backend — never the interface — is the enforcement authority.

05 Architecture

Built to be
audited, not just shipped.

A modular platform where permissions, tenancy and audit are enforced in the backend — so a new clinical module inherits the guarantees instead of re-implementing them.

VEHA Clinical Intelligence platform architecture Four client applications — an admin portal and clinical console on the web, and nurse and doctor mobile apps — talk to a REST API gateway. Behind it sit Spring Boot service modules, an AI service layer with guardrails, and an integration layer for hospital systems. These are backed by PostgreSQL as the system of record, Redis for sessions and throttling, object storage for attachments, and an observability stack for logs, metrics and traces. CLIENTS Admin Portalweb Clinical Consoleweb Nurse Appmobile · bedside Doctor Appmobile · review API Gateway REST · OpenAPI · OAuth2 / OIDC · rate limiting · tenant scoping SERVICES Platform Modules auth · users · roles · sessions clinical workflow · audit AI Services summarisation · drafting triage support · guardrails Integration Services FHIR R4 · HL7 v2 · DICOMweb HIS · EMR · LIS · PACS DATA & PLATFORM PostgreSQLsystem of record Redissessions · throttling Object Storageattachments · exports Observabilitylogs · metrics · traces Cross-cutting per-hospital tenant isolation · RBAC · append-only audit · encryption in transit and at rest · separate dev / QA / staging / production
Target architecture. Clinician-facing clients are thin; enforcement of permissions, tenancy and audit lives in the backend, never in the interface.

Why this runs in the cloud

Hospitals cannot each host an AI platform, and clinical data cannot sit on a single box. The workload needs managed Postgres with point-in-time recovery, isolated environments for development, QA, staging and production, GPU-backed inference for summarisation, object storage for attachments, a managed KMS for per-tenant key scoping, and an observability stack that can prove what happened during an incident.

It also needs to run in a region under a provider BAA, with selectable data residency — which is a cloud decision before it is an engineering one.

06 Interoperability

HL7 FHIR R4 is
the contract.

Everything else is normalised at the boundary, so core services only ever see one model — and a hospital is never asked to change its systems to adopt ours.

Hospital systems

  • FHIR R4 Patient, Encounter, Condition, Observation, DocumentReference
  • HL7 v2 ADT where a site has no FHIR endpoint
  • C-CDA and IHE XDS.b for discharge summaries and referrals

Diagnostics and imaging

  • FHIR ServiceRequest / DiagnosticReport, or HL7 v2 ORM/ORU over MLLP
  • DICOMweb (WADO-RS, QIDO-RS, STOW-RS) and FHIR ImagingStudy
  • No image copies stored by default

Devices and bulk data

  • Nurse-entered vitals first; device feeds mapped to FHIR Observation with LOINC codes
  • CSV and FHIR Bulk Data ($export, NDJSON) for sites without real-time integration
  • Streaming device data only after safety, scale and alert-ownership sign-off

Identity and terminology

  • OIDC and SAML 2.0 SSO, MFA, and SMART on FHIR app launch
  • SNOMED CT, LOINC, ICD-10/11, RxNorm and CPT mapped per site
  • Regional profiles plug in at the integration layer, not in core services

HL7 FHIR R4US Core / IPSSMART on FHIRHL7 v2.xC-CDADICOMwebIHE PIX/PDQ, XDS.b, ATNASNOMED CTLOINCICD-10

07 Security and compliance

Least privilege,
isolation, evidence.

HIPAA first, with GDPR and ISO controls built into the same platform rather than bolted on for an audit.

Identity and sessions

OAuth2/OIDC, short-lived access tokens, rotating refresh tokens, secure mobile token storage, MFA-ready admin access and forced logout on risk events.

Tenant isolation

Every user, patient workflow, audit event, session, role, report and configuration record is scoped to its hospital, region or unit.

Data protection

TLS 1.2+ in transit and AES-256 at rest across database, object storage, backups and exports, with keys in a managed KMS and no ePHI in logs, traces or URLs.

Audit coverage

Login, access denial, user lifecycle, role change, session revocation, clinical approval, AI draft change and consultation actions are all append-only events.

HIPAA / HITECHGDPRISO/IEC 27001ISO/IEC 27701ISO/IEC 27799SOC 2 Type IIHITRUST CSFNIST 800-53OWASP ASVS L2

Certification status. The frameworks above are design targets and control commitments. SOC 2 Type II and HITRUST require an observation window and an external auditor and are scheduled work — nothing here should be read as a claim of current certification.

08 Questions

Frequently
asked.

In development. VEHA Clinical Intelligence is not available for clinical use and nothing on this page should be read as a claim of regulatory clearance or completed certification. The security and compliance frameworks listed are design targets and control commitments; audited certifications such as SOC 2 Type II and ISO 27001 require an observation window and an external auditor, and are scheduled work, not current status.

What is VEHA Clinical Intelligence?

An AI-assisted clinical operations platform for hospitals and specialist care teams, covering bedside intake, AI-drafted documentation held under doctor approval, and Tele-ICU specialist collaboration — across web and mobile.

Does the AI make clinical decisions?

No. The governing rule of the platform is that AI produces draft support only — the medical decision, the approval and the accountability remain with an authorised clinician. AI output is labelled, editable and traceable, and escalation suggestions stay advisory until a clinician acts on them.

Is it available today?

Not yet. Clinical Intelligence is in active development and is not available for clinical use. We are talking to hospitals interested in shaping it and in early pilot participation.

How does it connect to our existing hospital systems?

HL7 FHIR R4 is the primary integration contract, covering Patient, Encounter, Condition, Observation and DocumentReference. Where a site has no FHIR endpoint we accept HL7 v2 (ADT, ORU, ORM) through an interface engine, DICOMweb for imaging, and C-CDA for documents. Sites with no real-time integration can start with CSV or FHIR Bulk Data import.

What about patient data and compliance?

The platform is being designed to HIPAA and GDPR from the start: TLS 1.2+ in transit, AES-256 at rest, managed KMS key handling, per-tenant scoping, append-only audit, and no ePHI in logs, traces or URLs. These are design commitments — see the status note on this page for what is and is not certified today.

Who is it for?

Hospitals running ICUs and wards that carry a heavy documentation load, and specialist networks that want to extend intensivist coverage to sites that cannot staff it around the clock.

09 Get involved

Shape it before
it ships.

We are looking for hospitals and ICU teams willing to pressure-test the workflow, and for clinical and engineering partners who want a say in how AI enters the medical record. Early pilot conversations are open now.