The platform

Not a chatbot bolted onto a hospital.
An operating system for one.

HealthTeky is a complete hospital management system — appointments, records, pharmacy, billing, wards — with an AI layer of 34 capabilities running across 10 departments on top of it. The AI is not a feature in a corner. It reads the same record the doctor reads, acts inside the same permissions, and stops at the same place every time: a human signature.

34
AI capabilities, shipped and tested
10
Departments covered, front door to audit
6
Agents working without being asked
65
Clinical evaluation cases it must keep passing

The problem it is built for

Hospitals do not lack data. They lack attention.

The record is too long to read

A patient with three admissions and forty lab results has a history no one has time to read before a seven-minute consultation. So it is not read, and the same question gets asked again.

Deterioration is visible in hindsight

The observations that predicted the arrest were all recorded. They were recorded on four different charts, by three different people, across eleven hours.

The safe thing depends on who is on shift

An interaction caught on Tuesday is missed on Sunday. Not because anyone is careless, but because the check lives in a person's memory rather than in the system.

How it is built

Four layers. The interesting one is the third.

Layer 1

The hospital management system

Patients, appointments, consultations, prescriptions, pharmacy, lab, billing, wards. Multi-tenant, so a group runs every hospital in one place with data separated per tenant.

Layer 2

The clinical engines

NEWS2, ESI, PHQ-9, GAD-7, EPDS, KDIGO, WHO growth standards, NTEP regimens, antimicrobial rules. Deterministic, published, and testable — they give the same answer twice.

Layer 3

The Guardian

Every AI action is classified into one of 5 levels and checked against 21 governed action types before it runs. It decides what needs a signature, from whom, and refuses the rest. Nothing reaches a patient around it.

Layer 4

The language model

Used for what it is good at — reading, drafting, summarising, translating — and never for arithmetic that a rule can do or a decision a clinician must make. It is the last layer, not the product.

What happens when a patient arrives

One patient's path, and what the AI does at each step.

1

Before they arrive

Pre-visit history taken on their own phone, in their own language, stopping on a red flag.

2

At the front door

Symptom text scored for acuity, with the phrase that caused it shown to the triage nurse.

3

In the room

The consultation dictated; a SOAP draft and safety-checked orders returned for signature.

4

On the ward

Every observation scored; deterioration escalated with a handover written for the receiver.

5

After discharge

Check-ins scored green, amber, red — the red ones back on a nurse's list the same day.

Every capability, by department

34 of them. 25 carry a named safety rule that decides what the AI may not do.

Start here 2

The AI HubOne tile per capability, and the four numbers that say whether the hospital is safe right now.
NotificationsWhat has happened since you last looked.

The record 2

Patient 360The whole record in about a minute, with every claim cited.
Ask the RecordAsk a question about one patient and get a cited answer, or an honest "not recorded".

Front door: triage, intake, emergency 3

Red-flag triageSymptom text in, acuity out, with the phrase that caused it.
Pre-visit intakeA history taken before the consultation, in the patient's language, that stops on a red flag.
Emergency department boardESI triage in its published order, ambulance pre-alerts, and a crowding score.

In the consultation 3

Ambient scribeDictate the consultation, get a SOAP draft and safety-checked orders to sign.
The approvals queueEverything the AI has prepared, waiting for a human signature.
Mental health screeningPHQ-9, GAD-7 and EPDS, where the self-harm item beats the total.

Ward, ICU and theatre 4

Nurse worklistWho needs you now, ranked, with an SBAR handover per patient.
Wards and dischargeBed board, admissions, discharge blockers and a drafted discharge summary.
ICU boardDeterioration trend, sepsis screening and Hour-1 bundle timers.
Surgery and the WHO checklistPre-operative risk, medicines to hold, and a checklist that blocks.

Medicines 3

Prescription safety checkAllergy, interaction, duplication, renal, pregnancy and age checks before signing.
PharmacyDispense-time re-check, first-expiry-first-out batches, stock cover.
Antimicrobial stewardshipThe round: what to narrow, what to switch to oral, what should have stopped.

Long-term and population 3

Chronic disease controlIs it actually controlled, and what to change - prepared for a prescriber to sign.
Population healthWho to call this week, control rates per register, and the gap a headline hides.
Syndromic surveillanceDoes today look unusual, and where.

Programmes 5

Mother and childWHO antenatal contacts, pre-eclampsia, danger signs, growth and immunisation.
TuberculosisNTEP notification, regimen, the month-5 rule, contacts and cohort outcomes.
Prevention and recallsScreening, chronic-care and vaccination gaps, each with the rule behind it.
Programme registersNCD, TB and immunisation returns, with the missing fields as a worklist.
Post-discharge recoveryCheck-ins scored green, amber or red, with free text read for red flags.

Governance and assurance 5

Guardian consoleWhat is deployed, what it may do, what it did, and whether it still passes its tests.
Reliability, drift and costScored evaluation runs kept over time, drift findings, and what the models cost.
AgentOSEvery agent decision logged and replayable, scheduled monitors, and plans approved once.
Command centreLive flow, occupancy, clinical risk and AI-safety telemetry in one view.
Revenue integrityDocumentation gaps that cause denials, with ICD-10 suggestions.

Interoperability and rights 4

Connect: FHIR and HL7FHIR R4 reads and HL7 v2 intake, with every message logged.
Consent and ABDMConsent artefacts with expiry and withdrawal, and ABHA identifiers.
Patient rights (DPDP)Access, correction, erasure and withdrawal requests, tracked to their statutory date.
Compliance packThe evidence an auditor asks for, assembled from what actually happened.
Every number on this page is counted from the running product. The capability list is the same list the software reads to build its own onboarding guide, so it cannot describe a feature that is not there.

Walk through it on your own data

We will load a training tenant with your case mix and take you through the 34 capabilities in the order a patient meets them.

Talk to us about Enterprise See all plans