Administrative first, clinical only with evidence
AI for hospitals and health systems on a hard line between administrative work and clinical judgement, with consent and residency handled properly.
The boundary
The prohibition list is written before the feature list and goes into the guardrails, the evaluation suite and the contract: no diagnosis, no severity assessment, no treatment advice, no result interpretation, and never that someone need not be seen.
Access
Booking, rescheduling, clinic routing by referral type, preparation instructions and reminders, on WhatsApp, web and voice in the languages the catchment speaks.
Safety
100%
of red-flag cases must reach a human or emergency pathway — a release gate
Standards
Integration against the hospital information system through FHIR, HL7 v2 and DICOM, so the record stays the record.
Documentation
Clinic letters, discharge summaries and structured fields drafted from the encounter. Every statement traceable; fabricated findings are the failure mode we test hardest for.
Data
Consent and purpose modelled as data and enforced at retrieval, minimisation designed in, and disconnected deployment with no external inference where the rules require it.
Standards
Systems
Regulation
Assurance
Deployment
3–5 weeks
Which workflows sit clearly on the administrative side of the line, and the classification question answered explicitly.
Safety boundary specification, hazard log, consent and residency constraints
10–16 weeks
One access or documentation workflow built end to end, silent-run against live traffic before a supervised release.
Working workflow, clinician-reviewed evaluation sets, silent-running results
Ongoing
Rollout across departments and sites, operation inside your environment under an SLA, with standing safety review.
SLA operation, safety review cycle, governance evidence, handover to hospital IT
Administrative work moves a patient through a pathway: booking, routing by referral type, preparation, reminders, correspondence a clinician signs. Clinical work forms a judgement about a person's health. We build the first by default; the second needs governance sign-off and prospective validation.
Yes, and for ministry, military and national environments it is usually the requirement: open-weight models on hospital infrastructure or a national health cloud, including disconnected sites. We have delivered into the Türkiye Ministry of Health, Bilkent City Hospital and Azerbaijan Military Hospital.
When the pathway itself is the problem. If clinic templates are wrong or the patient administration system holds duplicate records, an assistant will book people into the wrong clinic faster and more politely. And without clinician time to review evaluation sets and sit on the safety review, nothing can be validated.
Regulator, budget, deadline, legacy core, a board that has been burned once already. Tell us what you are working around.
