The AI workforce

Nine roles. Every shift.
Not one of them signs anything.

The easiest way to understand HealthTeky is to stop thinking of it as software and start thinking of it as staff — a team that never goes off shift, never forgets the interaction check, and never gets to make the decision. Each role below is a way of talking about features that actually ship; the links go to what implements it.

AI Doctor

Listens to the consultation and writes the note.

Dictate the consultation. It returns a structured SOAP draft with the orders and prescriptions extracted, each one safety-checked and queued for the doctor to sign. The doctor signs; it never does.

Backed by
Ambient scribeThe approvals queueAsk the Record

AI Nurse

Monitors every observation and raises the alarm.

Every set of vitals recorded anywhere in the hospital is scored on NEWS2 as it is saved, ranked into a worklist of who needs attention first, and escalated with an SBAR handover when it crosses the line.

Backed by
Nurse worklistWards and dischargeICU board

AI Pharmacist

Checks every prescription before it reaches a patient.

Allergy, interaction, duplication, renal dosing, pregnancy and age checks at the point of prescribing, again at dispensing, plus an antimicrobial stewardship round on what should be narrowed or stopped.

Backed by
Prescription safety checkPharmacyAntimicrobial stewardship

AI Triage Officer

Never lets a red flag walk back out of the door.

Symptom text in, acuity out, with the exact phrase that caused it. Pre-visit history taken in the patient's own language that stops dead on a red flag, and an ED board running ESI in its published order.

Backed by
Red-flag triagePre-visit intakeEmergency department board

AI Intensivist

Watches the ICU when the unit is thinnest.

Deterioration trends, sepsis screening and Hour-1 bundle timers, so the three in the morning handover is a screen rather than a memory.

Backed by
ICU boardSurgery and the WHO checklist

AI Discharge Planner

Finds the blocker before the bed is wasted.

Bed board and discharge blockers surfaced early, a drafted discharge summary in the patient's language, and post-discharge check-ins scored green, amber or red with free text read for red flags.

Backed by
Wards and dischargePost-discharge recovery

AI Epidemiologist

Sees the outbreak while it is still small.

Syndromic surveillance asking whether today looks unusual and where, control rates per disease register, and the care gaps a headline number hides.

Backed by
Syndromic surveillancePopulation healthPrevention and recalls

AI Coder

Stops the denial before the claim is sent.

Documentation gaps that cause denials, surfaced while the patient is still in front of you, with ICD-10 suggestions a coder confirms.

Backed by
Revenue integrityProgramme registers

AI Auditor

Can prove what happened, years later.

Every AI action and every refusal written to a hash-chained trail, replayable against the rules as they stood that day, and assembled into the evidence pack an accreditation visit asks for.

Backed by
Guardian consoleCompliance packAgentOS

Where every one of them stops

The Guardian sorts each action into one of these levels before it runs. The level decides whether a human must sign, and who is allowed to. This is not a setting a hospital can switch off to go faster.

ObserveReads and records. Cannot change anything at all.
SuggestOffers a draft to a human. The draft does nothing until someone acts on it.
PrepareBuilds a prescription or order and holds it in a queue. Releases only on a named clinician's signature.
ExecuteLow-risk administrative work — booking a follow-up, sending a reminder. No clinical effect.
RegulatedAnything a regulator governs. Requires the specific permission that law or policy attaches to it.

21 action types are governed this way, across 13 staff roles. Every refusal is recorded as carefully as every action.

What that means in practice

The AI cannot prescribe

It can draft a prescription, check it against allergy, interaction, renal function and pregnancy, and put it in a queue. A doctor signs it. If no one signs, nothing happens.

A nurse cannot approve a drug

Permissions are checked per action, not per screen. A multi-step plan requires every permission its steps need — not the one belonging to the biggest step.

You can replay any decision

Each action is written to a hash-chained trail with the inputs it saw and the rules in force that day, so a question asked two years later has an answer.

Meet the workforce on your own ward

We will run the AI Nurse against a week of your real observations on a training tenant, and show you what it would have escalated and when.

Talk to us about Enterprise See all plans