Healthtek
Vishnu Dental College & Hospital
Everything the incumbents do — plus the AI they bolt on.
A sovereign, AI-native EMR + EHR with an ambient voice scribe — in any language — that runs entirely on one on-prem appliance: private, safe patient care with 99.9%-class uptime, and no patient data ever leaving the hospital. For a teaching hospital it carries a second half no EMR does — a dental college layer for students, faculty and DCI requirements, built from the same clinical rules (page 3).
Patient data access
Unified, UHID-rooted record; instant chart on the iPad — served on-prem, no cloud round-trip.
Ambient AI documentation
Voice → SOAP in ~2.5 s, in any language or code-mix, e-signed before the patient leaves the chair.
Appointment scheduling
Front-desk + patient self-booking portal with a live slot grid — fewer no-shows, smoother OPD.
Telehealth services
Integrated video consults — one secure link per visit, auto-logged to the chart with duration.
Patient engagement
UHID+PIN portal (records, Rx, labs, SOAP PDF) with WhatsApp / email delivery.4
Medication management
e-Prescribing + stock-aware pharmacy + billing & claims in one flow.
Analytics & reporting
Live-Ops console + per-doctor activity, computed on the box — no data export.
Sovereign local AI
Local models on-device; PII-scrubbed before any optional cloud LLM — PHI never leaves.
The spec & comparison sheet
| Layer | Capabilities |
|---|---|
| Ambient AI Scribe | STT — any language via best-in-class engines: Sarvam · Bhashini · Deepgram · Grok, with an on-device offline fallback.6 Note-gen cascade: local Claude → Anthropic API → local Ollama gemma4 → OpenRouter. PII-scrub (spaCy NER) before any cloud call · >30 s auto-chunking · department-aware SOAP for any speciality · ~2.5 s · e-signed PDF. |
| Clinical / HIMS | OPD · IPD · OT · e-prescribing · vitals · labs + OCR · pharmacy + live stock · billing / RCM (claims · tax · bank-file) · ANC / pregnancy · appointments · medical-knowledge tile · FHIR/HL7 interop on the roadmap.5 |
| Patient & Telehealth | UHID + PIN patient portal (records · Rx · labs · downloadable SOAP PDF) · self-service booking · integrated LiveKit video consults (one-link-per-visit, auto-logged) · WhatsApp / email delivery.4 |
| Platform & Security | On-prem AI appliance — NVIDIA DGX-class, 128 GB unified memory, ~4 TB.7 Tailscale mesh · PIN-based RBAC · Postgres 16 + pgvector · AES-256 encrypted nightly backups · full data portability (no lock-in) · one strictly-allowlisted public portal ingress. |
| Deploy & Evolve | One appliance per ~20 doctors · flash-image install in hours · unattended, reboot-safe · 99.9%-class uptime · recursive-learning agents + live ticket resolution keep it bug-free and evolve it to each site. |
Healthtek vs the enterprise incumbents
| What it really takes | Healthtek · Care in a Box | Epic | Oracle Health | Legacy Cloud EHR |
|---|---|---|---|---|
| Licence & setup | One appliance + a small annual contract. | Multi-million licence; implementation often 2–3× the licence. | Multi-million enterprise licence + services. | Per-seat cloud fees that compound every year. |
| Go-live time | Weeks — flash-image install in hours. | Quarters to years, with large on-site teams. | Multi-year rollouts. | Months + data migration. |
| Support model | Live ticket resolution + AI agents — no on-site army. | On-site specialists + paid professional-services retainers. | Vendor PS + tiered support contracts. | Ticketed cloud support. |
| Maintenance & upgrades | Self-maintaining; agents keep it bug-free. | Scheduled upgrades + change control + ongoing hosting fees. | Managed-cloud maintenance fees. | Vendor-controlled release cycles. |
| Gets better with use | AI evolves to your workflows — the more you use it, the better it fits. | Static until the next paid upgrade. | Vendor roadmap only. | Vendor roadmap only. |
| Where data lives | On your own box — PHI never leaves. | Epic / cloud-hosted. | Oracle Cloud. | Vendor cloud. |
| Ambient AI & languages | Native, on-box, any language & speciality. | Cloud add-on (DAX / Abridge), extra per-provider fee; English-first. | Cloud add-on; English-first. | Usually none / third-party. |
| 5-year cost, indicative | A fraction — appliance + small AMC. | $ millions. | $ millions. | Compounding subscription. |
The half a hospital EMR doesn't have.
A dental college is two institutions sharing one building: a hospital that treats patients, and a college that has to prove its students are fit to. Every EMR serves the first. This one serves both — from the same clinical rules, on the same box. Students train against the identical logic that raises the contraindication banner at the chair, so a drill can never teach an answer the live system would not give.
The six drills
| Drill | What it asks, and what marks it |
|---|---|
| Contraindication | 60 cases. A patient's medications, conditions and allergies against a planned procedure — what is the risk, and how serious. Marked by the same rules engine that raises the chair-side banner, resolved live at commit time, so a corrected rule corrects the drill. |
| Diagnosis | 60 cases. Pulpal and apical diagnosis from structured findings, in AAE terminology. Marked on the two axes separately — different tests, different blind spots. Refuses to name a diagnosis for contradictory findings rather than guessing. |
| Charting | 40 cases. "Upper right first molar" → put it on the chart in FDI/ISO-3950. Errors are reported by quadrant, because that is how FDI mistakes cluster. Opens a year earlier than the rest — it needs no patient. |
| Explaining treatment | 13 cases. Explain the treatment to the patient in their own language. Marks only what rules can check — an over-promise of "no pain at all", a drug named with a dose — and says so on screen. The reference answer is labelled "one good answer, not the answer". |
| Chair-side emergency | 10 cases. Syncope, angina, anaphylaxis and the rest — what you do and in what order. Built on Resuscitation Council (UK) sequences, marked on sequence, not recall. The instinctive answer is usually the wrong one.10 |
| Virtual patient | 6 cases. Take a history from someone who will not volunteer it. A model plays the persona; the facts are authored data and the marking is set comparison. Ask four questions without mentioning medication and the patient will chat happily and never mention the drug that matters. That is the lesson. |
Faculty, quota and countersignature
Verified countersignature
A countersignature is the college's record that a qualified teacher supervised a student on a real patient. The signer is the session, not a typed name — resolved to a person, with their teaching rank.
DCI quota tracking
Ten procedure categories per student, counted only once a teacher has signed. One screen answers "is this student ready?" — clinical work and assessment side by side.
Faculty write their own cases
The college owns its content. Faculty author cases for three drills, including the virtual patient's persona, what it admits and when. No edit after answers exist — withdraw and rewrite.
What they keep getting wrong
Read out of the student's own answers, on the axis that teaches: the quadrant, the pulpal vs apical split, the question they are not asking. Counts, never percentages.
Pre-clinical log
Phantom-head and typodont work, with a photo. Graded by nothing — assessing a wax carving from a photograph is something a model does confidently and badly. It waits for a tutor.
Marks are immutable
An attempt is never edited; a repeat adds a row. A case with answers against it is never edited either — changing it moves the ground under a mark already given.
Year-gated, not role-gated
Capability follows academic stage, derived from year rather than stored — so a student cannot hold a privilege from a year they have left, and a locked drill says which year it opens.
Own the whole record
Drills, bench work, clinical quota and faculty grades sit in the college's own Postgres, on the college's own box. Exportable, inspectable, no vendor in the middle.
Against an LMS, and against a plain EMR
| What a college needs | Healthtek · the college layer | Moodle / an LMS | A hospital EMR |
|---|---|---|---|
| Anchored to real clinical work | Same rules engine as the chair-side banner; quota counted from real logged procedures. | Works perfectly with an empty patient table — it cannot know. | Records the treatment, knows nothing about the student. |
| Who marked this | Rules a tutor can read, plus a named human grade — stored in separate columns so they are never confused. | A quiz score, or a rubric a human filled in. | — |
| Countersignature | Resolved to a person and their teaching rank; not editable once given. | — | A free-text name, if anything. |
| DCI quota readiness | Ten categories, live, per student, from signed work. | A spreadsheet somebody maintains. | — |
| History-taking practice | A patient who will not volunteer the drug that matters — and marks the question you failed to ask. | Multiple choice. | — |
| One identity | One login for the student's coursework and their clinical record. | A separate account and a separate world. | Staff accounts only. |
Two minutes, every feature — live.
• Administrator — Steven Kane / 4455 — full access: every department, RCM Intelligence, Live Ops (doctor-usage analytics), billing & finance.
• Billing & Finance — Karen Whitfield / 5566 — the money desk: billing, collections, dues, payouts, pharmacy & RCM claims — no clinical charts.
• Doctor — Dr. Anjali Prasad / 1234 — clinical only: EMR, voice scribe & SOAP, labs, Rx, the odontogram & department pages — no billing / RCM.
• Faculty — Dr. K. Srinivas / 1234 — the teaching view: review queue, student roster, case bank, and countersigning clinical work.
• Student — roll number 21-BDS-114 / 1234 — the six drills, My Progress, and her own clinical case-log.
Patient portal (vdcapp.deepromallick.com/portal) — demo logins (UHID / PIN):
• Anjali Sharma — VDC-DEMO-OBS / 222222 — obstetrics; a full antenatal history (SOAP, Rx, labs) in her chart.
• Ramesh Naidu — VDC-DEMO-GEN / 555555 — general medicine (hypertension follow-up).
• Aarav Kumar — VDC-DEMO-PED / 111111 — paediatrics.
• Sunita Patel — VDC-DEMO-DRM / 444444 — dermatology.
• Priya Iyer — VDC-DEMO-FRT / 333333 — fertility.
Synthetic demo patients only — no real patient data.