Specialist care that travels through cameras, smart glasses and the cloud. This deck walks through our understanding, the technical stack, the architecture, the development plan, and a clear cost picture.
Prepared by Tiruveedula Jagadeesh
v1.0 · 9 May 2026 · Hyderabad, India
A 58‑year‑old farmer is wheeled into a 30‑bed hospital in Anantapur. His oxygen is dropping. The duty doctor is a brilliant generalist — but the nearest intensivist is 180 km away. The next four hours will decide everything. This is the moment Veha was built for.
Onboards hospitals, assigns doctor pool, owns billing & platform health.
Adds wards, beds, devices and staff. Manages shifts and reports.
Watches live, drives PTZ, talks via glasses, signs the discharge summary.
Wears glasses, admits patient, executes the doctor's instructions.
Out of MVP: 3rd‑party EMR connectors, mobile apps for staff/doctor (deferred).
Node.js + NestJS (TypeScript) — same language across BE/FE. Spring Boot acceptable if team is Java‑heavy.
REST + WebSocket (PTZ, alerts, presence).
PostgreSQL on Azure Flexible Server (Burstable B1ms). JSONB for flexible clinical fields.
Redis Basic C0 for sessions & presence.
React 18 + Vite + TypeScript + TailwindCSS + shadcn/ui. Doctor console + Admin portals.
No mobile app in MVP.
LiveKit (self‑hosted, 1 small VM) for WebRTC. go2rtc / MediaMTX on edge gateway converts RTSP→WebRTC.
Container Apps (scale‑to‑1), Blob Storage (Hot→Cool→Archive), Front Door + WAF, Key Vault.
Azure Speech (VAD‑gated STT) + Azure OpenAI gpt‑4o‑mini for discharge summaries & "ask the chart".
Build time: Copilot Business + Cursor + ChatGPT Team — projected ~25% engineering compression.
Mini PC on hospital LAN. Runs go2rtc/MediaMTX + buffer. Cameras never face the public internet; the gateway pushes outbound TLS to LiveKit.
JWT + email OTP, MFA for doctor/admin, per‑bed ACL, full PHI audit log.
Hospital → ward → bed → device hierarchy. Schema‑per‑tenant‑ready.
PTZ + glasses + vitals + transcript on one screen. Multi‑doctor view, single PTZ control token.
English + 1 Indian language, speaker diarization, searchable transcripts.
gpt‑4o‑mini drafts SOAP discharge summary, doctor edits & signs.
SOS, "code blue" keyword, device offline, doctor‑raised. Email + SMS + browser push.
Continuous while admitted. Hot 7 d → Cool 30 d → Archive.
Admit → vitals → notes → orders → discharge. PDF + FHIR export.
Occupancy, response time, alert volumes, per‑tenant cost.
99.5% on media plane in MVP.
≤ 800 ms audio RTT on glasses.
Under typical pilot load.
TLS 1.2+ in transit; signed URLs for media.
Append‑only; tamper‑evident hashes on recordings.
DPDP‑aligned, HIPAA‑ready, ISO 27001 path.
≈ 26 weeks from kickoff to first hospital go‑live, with an AI‑augmented team of ~5.5 FTE.
×1 · Owns BE + FE + Azure.
×2 · Node + React, AI‑augmented.
×1 · LiveKit, ONVIF, glasses.
×1 · Playwright + GitHub Actions.
×0.5 · Figma + shadcn.
×0.5 · Workflow validation.
≈ 25% velocity boost.
~5.5 billable.
Server & tools to build the platform once. Reused for every future hospital. Development effort billed separately.
~₹1.45 L6 months · server + tools only
Everything the hospital pays for, in two parts:
Doctor fees, support staff and clinical costs sit outside these buckets and feed the subscription price directly.
Lean setup. Postgres runs on a single self‑managed VM, no managed databases, no Media Services, no AKS during build. Founder / engineering salaries are absorbed by the team and excluded. We upgrade infra only when real app usage demands it.
| Line item | Detail | ₹ (6 mo) |
|---|---|---|
| Single VM (dev + staging) | 1× Hetzner / Contabo CX22, Docker, self‑managed Postgres, Nginx, MinIO · 6 mo × ₹1,500 | 9,000 |
| Object / blob storage | MinIO on same VM · ₹500/mo backup snapshot to B2 / Wasabi | 3,000 |
| Domain + SSL | .in domain · Let's Encrypt SSL (auto‑renew) | 2,000 |
| Email / transactional | Resend / Brevo starter · paid SMS OTP top‑up | 3,000 |
| Observability | Self‑hosted Grafana + Loki on same VM · Sentry starter | 0 |
| AI coding tooling | Copilot Pro+ × 3 seats · 6 mo @ $39 ≈ ₹3,315/seat/mo | 59,670 |
| Design + repo | Figma starter, GitHub · 1 paid seat | 5,000 |
| Pen‑test (lite, before pilot) | One‑off external review before go‑live | 50,000 |
| Contingency (10%) | 13,170 | |
| Total Bucket A (server + tooling, 6 mo) | ~₹1,45,000 |
Upgrade path (only when real usage shows up): single VM ➜ managed Postgres on Azure / DO ➜ AKS + Media Services ➜ multi‑region. We pay for managed cloud only after the first paying hospital is live.
| What we pay for | ₹ / hr |
|---|---|
| Saving the camera recording | 0.5 |
| Streaming live video to the doctor | 5.0 |
| Video room (doctor ↔ bedside nurse ↔ ICU team) | 5.5 |
| Voice‑to‑text for nurse rounds (~10 min) | 18.0 |
| Auto handoff summary (every 3 days) | 1.5 |
| Shared platform cost | 4.5 |
| Safety buffer (15%) | 5.0 |
| Total per active bed‑hour | ~₹40 |
| What’s happening at the bed | ₹ / hr |
|---|---|
| Bed quiet — only camera recording | 6 |
| Doctor watching the live video | 15 |
| Nurse on glasses, dictating notes | 27 |
| Full active encounter | 40 |
| What we pay for | What it does for the hospital | ₹ / mo |
|---|---|---|
| App servers | Run the doctor / nurse screens, login, alerts | 2,900 |
| Live video room | Carries doctor ↔ patient ↔ nurse video calls | 2,490 |
| Patient + vitals database | Stores every reading, note and alert, with daily backup | 2,075 |
| Fast memory cache | Keeps screens snappy under load | 1,410 |
| Recent recordings (last 7 days) | Fast playback for any bed in the last week | 250 |
| Older recordings (up to 1 month) | Slower but cheaper storage for last month | 500 |
| Long‑term archive | Compliance archive of everything older | 915 |
| Internet bandwidth | Sending video out to doctors’ phones / laptops | 1,825 |
| Voice‑to‑text for nurse rounds | ~14,400 minutes of dictation per month, auto‑typed | 11,950 |
| AI handoff summary | 72‑hour shift summary written for the next doctor | 85 |
| Email + OTP SMS | Login OTPs, escalation alerts to on‑call doctor / ICU lead | 580 |
| Security, login, audit, monitoring | Firewall, single sign‑on, audit log, error alerts | 2,905 |
| Backup retention | Long‑term legal retention of backups | 830 |
| Safety buffer (~15%) | Reserve for spikes | 4,320 |
| Total | ~₹6,650 / bed / mo | ~33,000 |
Same software, same features. The only thing that changes is whether your hospital sits inside a shared building or in its own private one.
All hospitals run on the same Veha cloud. Each hospital's data is locked behind its own login and access rules — no hospital can see another's data.
Your hospital gets its own private cloud space — its own database, its own video room, its own storage. Nothing is shared with other hospitals.
TLS 1.2+ in transit, AES‑256 at rest, signed URLs for media.
JWT + email OTP, MFA mandatory for SuperAdmin & Doctor, per‑bed RBAC.
Azure South India only. RA‑GRS for archive.
Append‑only PHI access log retained 7 years; tamper‑evident hashes on recordings.
DPDP‑aligned, HIPAA‑ready controls, ISO 27001 path post‑GA.
PHI redaction in prompts, doctor‑in‑the‑loop on every signed artefact.
| Risk | Mitigation |
|---|---|
| Hospital Wi‑Fi unreliable | Edge GW buffers locally; degrade to 480p / audio‑only. |
| Glasses battery & heat in long shifts | 2 spare batteries per glass, 4‑hour rotation policy. |
| ONVIF PTZ quirks across vendors | Lock to one certified model in MVP; abstract driver layer. |
| Cloud cost surprises (egress, STT) | Per‑tenant cost dashboard + monthly soft caps + alerts. |
| Compliance gaps (DPDP/HIPAA) | Phase‑0 gap analysis; pilot only under signed BAA + DPA. |
| AI summary hallucination | Doctor edits & e‑signs; transcript + vitals shown side‑by‑side; PHI redaction. |
720p / 1 Mbps for ICU view — doctor‑clear, ~55% lighter on storage than full HD.
Transcription runs only when the nurse actually speaks — ~50% saving.
We host the video bridge ourselves — flat monthly cost instead of pay‑per‑minute.
Last 7 days fast, last 30 days slower & cheaper, older as cold archive — 90%+ saving on long‑term.
We use a smaller AI for shift summaries — about 30× cheaper than the top‑tier model, same quality for our use.
Off‑hours, the app servers shrink to a single small instance — we don’t pay for idle capacity.
6 months · development cost billed separately
5 beds · 2 worn devices · install · incl. GST
~₹6,650 / bed / month
full encounter mode
~6 months · Phase 0 → hardening
Save lives where intensivists can't physically be.
Once we have your answers, we lock the scope, the price and the go‑live date in writing — one page, no jargon.
Prepared by Tiruveedula Jagadeesh
v1.0 · 9 May 2026 · Hyderabad, India