TBYTE.dev · Production-First Healthcare Engineering

ICUVista

Critical Care, Connected Everywhere.

A secure, AI-assisted Tele-ICU clinical collaboration & command-centre platform — bringing specialist critical care to every ICU bed in India, over web, mobile and AR smart glasses.

Stakeholder Presentation July 2026 Baseline v1.0 · Approved

🩺 Specialist care, within reach

The Problem · भारत की चुनौती

India's critical-care capacity is stretched to breaking point

Critical-care expertise is scarce and concentrated in metro cities, while patients in district and rural hospitals face life-threatening delays in specialist intervention.

3,00,000+

ICU beds across India (ISCCM estimate)

~5,000

Trained intensivists managing them — 1 for every ~60 beds

2.3 / 100k

ICU beds per 100,000 population — far below global standards

Metro-heavy

Specialists concentrated in tier-1/2 cities; district & rural ICUs left behind

Global Tele-ICU market (USD)

The Opportunity

Tele-ICU is the scalable answer — and India is ground zero

One specialist, many beds: remote intensivists cover scheduled rounds and 24×7 emergencies across multiple hospitals.
Proven model in India: leading groups (e.g. Medanta e-ICU, Apollo TeleHealth) are already scaling tele-ICU networks.
Government & enterprise demand: district-hospital digitisation and national digital-health programs open procurement channels.
Ayushman Bharat Digital Mission momentum makes standards-based (FHIR R4) platforms procurement-ready.

Our Vision

Extend specialist critical-care expertise to every authorized ICU bed

Through secure virtual rounds, emergency support, hands-free bedside collaboration and AI-assisted clinical documentation — while preserving bedside physician responsibility, privacy and full traceability.

Virtual Rounds

Scheduled & emergency specialist consultations, patient-bound and consented.

Hands-Free Bedside

AR Android smart glasses stream the bedside view while clinicians keep working.

AI Documentation

Transcription, summaries & monitor-image extraction — always human-validated.

Trust & Audit

Tenant isolation, consent, versioned signed records and append-only audit.

🇮🇳 India's specialists — connected to every ICU

Product Identity · Approved Baseline

One platform, three channels, enterprise-grade cloud

Product	ICUVista — “Critical Care, Connected Everywhere.”
Category	Secure AI-assisted Tele-ICU clinical collaboration & command-centre platform
Channels	Web · Android/iOS mobile · AR Android smart glasses
Cloud	Microsoft Azure (AKS, Service Bus, PostgreSQL, Blob, Front Door)
Engineering	TBYTE.dev — production-first healthcare engineering (tbyte.dev)
Status	v1.0 Approved Baseline — PRD · Architecture · Database · UML complete

ICUV-PRD-001

Product Requirements & Module Definition

ICUV-ARCH-001

Technology Specification & Enterprise Architecture

ICUV-DBD-001

Database Design & ER Diagrams

ICUV-UML-001

Actors, Use Cases, Sequences & Class Model

All four engineering baselines are approved for design & development — we are ready to build.

Core Principles · Non-Negotiables

Clinical safety and privacy are designed in, not bolted on

Bedside owns care

Treating physicians retain final bedside clinical responsibility — always.

AI is assistive

Every AI output stays a draft until a clinician reviews and approves it.

Layered access

Tenant, hospital, ICU, patient, role, shift, purpose and time validity — all enforced.

Zero identity leaks

No patient identity in room names, URLs, logs or lock-screen notifications.

Care never waits

Emergency bedside care is never delayed waiting for a remote specialist.

Immutable records

Signed records are versioned; corrections via addenda, never silent overwrites.

Built for real networks

Adaptive video, audio-only fallback, reconnection and queued uploads for Indian connectivity realities.

End-to-End Workflow

From hospital onboarding to signed discharge summary

1

Configure & Onboard

Subscription plan → hospital, ICU units, beds, users, rosters.

2

Admit Patient

Registration, consent capture, ICU bed assignment, care-team mapping.

3

Assign Specialists

Bedside clinicians + remote specialist coverage with shift validity.

4

Consult Securely

Scheduled round or emergency — from mobile, web or smart glasses.

5

Recommend & Decide

Specialist advises; bedside physician accepts, modifies or rejects.

6

Task & Track

Approved actions become auditable tasks for doctors, nurses, technicians.

7

AI Drafts

Consented audio/images → drafts with confidence & source evidence.

8

Close & Audit

Signed summary at discharge/transfer; full audit history retained.

Actors & Industry-Standard Roles

12 defined roles, each with a hard access boundary

Platform Super Admin

Plans, hospitals, security & governance — no default patient access.

Operations Coordinator

Rounds, emergency routing, SLA — cannot approve care.

Clinical Quality Reviewer

Documentation audit — cannot silently edit signed notes.

Hospital Administrator

Beds, users, rosters — clinical access only when granted.

Bedside Physician

Primary clinical ownership; decides on recommendations.

Duty Medical Officer

Shift-based assessment, escalation & handover.

Remote Specialist

Rounds & emergency support — only assigned patients.

ICU Nurse & Technician

Observations, monitor capture, tasks, device support.

Hospital Support Coordinator Billing & Finance Officer Security & Compliance Auditor + AR Glasses User · External EMR/HIS · AI Provider (system actors)

Platform Scope

18 functional modules covering the full clinical & commercial lifecycle

Identity & Access

MFA, roles, scopes, break-glass

Organization

Hospitals, ICUs, beds, config

Subscription & Billing

Plans, usage, invoices, taxes

Patient & Encounter

Admission, transfer, closure

Care-Team Assignment

Physician, nurse, specialist, shifts

Scheduling

Availability, rounds, on-call

Virtual Consultation

Secure WebRTC, multi-party

Smart Glasses

Hands-free bedside client

Emergency Escalation

Severity, SLA timers, routing

Clinical Record

Notes, labs, timeline, procedures

Recommendations & Tasks

Lifecycle, evidence, due times

AI Clinical Assistance

Speech, summary, OCR, review

Consent & Privacy

Versioned, guardian, withdrawal

Command Centre

Rounds, sessions, connectivity

Clinical Quality

Gaps, adherence, feedback

Reports & Analytics

Clinical, commercial, government

Audit & Compliance

Access, AI provenance, exports

Integration Hub

FHIR/HL7, labs, imaging, devices

Flagship Capability · UC-06

Secure, patient-bound virtual consultations

Authorization before media: identity, tenant, patient assignment, consent and consultation status verified before any token is issued.
Short-lived tokens, opaque rooms: room IDs are random — no patient or hospital names, ever.
Encrypted WebRTC (LiveKit) with adaptive bitrate, reconnect and audio-only fallback for low bandwidth.
Recording off by default — enabled only with contractual, policy and consent approval.
Call quality measured: join time, jitter, packet loss, reconnects — feeding SLA reports.

🔒 Encrypted, consented, fully audited

Differentiator · AR at the Bedside

Hands-free smart glasses — a first-class clinical client

Voice-driven workflow: patient confirmation, consultation start, camera framing, monitor capture, note dictation.
Scan-to-verify: bed/patient QR scan and identity confirmation before any clinical media is transmitted.
Visible privacy indicator whenever camera, mic, streaming or recording is active.
No independent data store: minimal encrypted cache, auto-purged after sync; same authorization as mobile/web.
Graceful fallback to a paired Android phone/tablet on limited input or network.

Emergency Escalation · UC-10

When seconds matter: SLA-timed specialist routing

A bedside clinician raises severity + reason + observations. The platform does the chasing — bedside care continues on local protocol, never blocked.

1

Raise Emergency

Reason, severity, current observations. SLA timer starts instantly.

2

Primary Specialist

Push/SMS/call notification; acknowledgement tracked against SLA.

3

Backup + Operations

No acknowledgement → automatic routing to backup specialist and ops centre.

4

Join & Resolve

Urgent consultation created; response & resolution times recorded.

Handles: no network · specialists busy · notification failure · duplicates Every escalation fully audited

🚨 Bedside care never waits

AI Clinical Assistance

AI that saves clinician time — under strict human governance

Speech → Text

Consultation transcription & dictated notes, incl. translation.

Summaries

Draft consultation & clinical summaries with source evidence.

Monitor OCR / Vision

Extracts vitals from monitor images with confidence scores.

Document Processing

Structured extraction from reports and attachments.

🛡️ Governance guarantees (ADR-009)

AI drafts can never auto-sign notes, orders or treatments — human approval is mandatory.
Full provenance on every output: model, version, prompt version, source, confidence, reviewer decision.
Patient data goes only to approved providers under contractual & regional controls.
Training data, evaluation sets and production data are strictly separated.
Provider abstraction: swap Azure-hosted / private models without touching clinical workflows.

Technology Stack · Approved (ADR-001…010)

Modern, proven, enterprise-ready

Experience Layer

Web: React + TypeScript, Vite, Material UI, TanStack Query, Zod
Mobile: React Native + Kotlin/Swift native modules
Glasses: native Kotlin AR client

Core Services

Java 21 LTS + Spring Boot microservices — Spring Security, JPA/Hibernate, Flyway, MapStruct, OpenAPI

AI Services

Python 3 + FastAPI async workers, model registry, evaluation, approved speech/vision/LLM providers

Data Layer

Azure PostgreSQL (system of record) · Managed Redis (cache/presence) · Blob Storage (encrypted media)

Real-Time Media

LiveKit / WebRTC, TURN/STUN, short-lived tokens, optional egress to Blob

Azure Platform

AKS + Helm · Service Bus · Front Door + WAF · API Management · Entra ID · Key Vault · OpenTelemetry + App Insights

DevSecOps: SAST · SCA · SBOM · container & IaC scans · signed releases 14 bounded microservice domains

Enterprise Architecture on Azure

Logical architecture — edge to AI, one governed platform

Front Door + WAF + APIM guard the public edge.
AKS private cluster — separate node pools for core, AI and media workloads.
Service Bus drives workflows, AI jobs & domain events (outbox pattern, idempotency).
Private endpoints for PostgreSQL, Redis, Blob, Key Vault.
Isolated environments: dev / test / UAT / production subscriptions.

Security, Privacy & Data Governance

Built for healthcare-grade trust — DPDP-era ready

RBAC + ABAC

Roles bound by tenant, hospital, patient, assignment, purpose & time. Denied by default.

Tenant Isolation

Every record carries tenant context; automated tests prove Hospital A can never read Hospital B.

Consent Engine

Versioned, guardian-aware, withdrawable — scoped to recording, transcription, images, sharing.

Append-Only Audit

Access, break-glass, exports, AI provenance — tamper-evident, SIEM-integrated.

Break-Glass

Reasoned, time-limited emergency access with instant security notification & review.

Encryption Everywhere

TLS in transit, encryption at rest, Key Vault secrets, malware-scanned private Blob storage.

Versioned Records

Signed notes keep prior versions + signer evidence; addenda, never overwrites.

Data Classification

Restricted clinical · sensitive identity · commercial · operational · security — each with matched controls & retention.

Non-Functional Baseline

Engineered to measurable commitments

99.9%

Production availability target (pilot 99.5%), excluding approved maintenance

<500 ms

Standard API p95 latency under agreed load

<2 s

Consultation media-token generation

RPO 15m / RTO 4h

Proposed enterprise disaster-recovery targets

Horizontal scale for APIs, AI workers and media nodes — new hospitals & regions without redesign.
Zone-redundant PostgreSQL, automated backups, tested point-in-time restore.
Reliability patterns: idempotent commands, retries, dead-letter queues, degraded-mode operation.
FHIR R4-compatible APIs — Patient, Encounter, Observation, Consent, Task, CarePlan and more.
Accessible, high-contrast clinical UI, responsive across web & mobile.
Secondary-region DR with periodic failover exercises.

MVP & Roadmap

A staged path from pilot to national scale

Phase 1 · Pilot MVP

Prove the core

Hospital / bed / patient setup
Assignments & scheduled rounds
Secure video/audio + notes
Recommendations → tasks
Consent, audit, basic subscription

Phase 2 · Clinical Intelligence

Amplify the bedside

Smart-glasses workflow
Monitor capture + AI drafts
Patient timeline
Emergency escalation & handover

Phase 3 · Enterprise Scale

Grow the network

Multi-hospital hierarchy
HA/DR & advanced reporting
FHIR integrations
Private deployment & governance

Phase 4 · Advanced AI

Assist, responsibly

Validated deterioration trends
Decision-support reminders
Controlled model improvement under clinical governance

ICUVista complements the hospital EMR — it does not replace it in release 1 No autonomous diagnosis or unsupervised prescribing — excluded by design

How We'll Measure Success

Clear KPIs across clinical, technical & commercial dimensions

Clinical Effectiveness

Virtual-round completion rate · specialist response time · recommendation acknowledgement & task completion rates

Platform Reliability

Successful call join rate · reconnect success · audio-only fallback rate under degraded networks

AI Quality

Transcription/summary correction rates · monitor-extraction confirmation accuracy

Security Posture

Unauthorized-access test pass rate · security-event closure time

Commercial Growth

Active hospitals · licensed beds · renewals · subscription utilization

Customer Delight

Hospital & clinician satisfaction, adoption depth per ICU

10 end-to-end test scenarios (E2E-01…10) already defined against these outcomes

The Ask · धन्यवाद

Every ICU bed in India deserves a specialist within reach.

The engineering baseline is approved. The market is proven and growing. We are ready to build the Pilot MVP.

1 · Endorse the baseline

Sign off the v1.0 PRD, architecture, database and UML baselines presented today.

2 · Nominate pilot hospitals

2–3 partner ICUs (mix of metro hub + district spoke) for the Pilot MVP.

3 · Approve pilot funding

Greenlight Phase-1 build, Azure environments and clinical governance setup.

ICUVista · Critical Care, Connected Everywhere.
TBYTE.dev — Production-First Healthcare Engineering · https://tbyte.dev