Ninkarrak — the Medical implementation of Baru, the trusted intelligence-fusion platform. In development. It is being built as a whole clinical visit in a station — multi-sensor screening and triage support, governed end to end, with the AI intended to run on-device and the clinical decision always human.
It screens and suggests; it never diagnoses.
Ninkarrak is being built as a self-contained smart health station: a patient will walk in, be identified and consented, have a full set of clinical-grade vitals collected, consult a registered physician over an encrypted link, and leave with a disposition — self-care, referral, or emergency. Every step is an Anshar zone with defined entry and exit conditions, so nothing advances until its preconditions are provably met.
The AI is designed to run on-device at the point of care — targeting a complete vitals sweep in under three minutes and a physician briefing in under fifteen seconds — with no protected health information leaving the station except under a lawful, logged disclosure.
Ninkarrak is being built as a self-contained, deployable screening station: multiple sensing modalities fused on-device into a single assessment that supports triage — in clinics, pharmacies, remote communities, and anywhere a station can be safely placed. It proposes a priority; a clinician decides. Like every Baru implementation, it works where the network doesn’t, on infrastructure its operator controls.
The design is built on two interlocked continuums — a patient-facing clinical session and a between-session operational cycle. They meet at one boundary: the clinical session is designed not to begin until the station has certified itself clean and ready — the interlock intended as the architecture’s primary safety mechanism. Every step is a zone with defined entry and exit conditions, so nothing advances until its preconditions are provably met.
Arrival → discharge: approach & access, identity and consent, sensor collection, live consultation, and one of three mutually-exclusive exits — self-care, referral, or emergency escalation.
Runs after every session: closure and safety, decontamination cycle, then readiness certification — the station vouches for itself before the next patient can enter.
Every zone is structured in three layers — Policy (what must happen), Conops (how it happens), Reporting (how it’s measured). AI lives only in the lower two, as sensors, actuators, and measurement synthesisers.
Runs on-device with no cloud round-trip — vitals, computer vision, acoustics, occupancy fusion, and sensor self-calibration. Structured, auditable outputs enter the zone as conops-layer inputs.
LLM synthesis for narrative output — plain-language consent, symptom transcription, clinical-note and referral drafts. Always conops-layer, always reviewed by a human before any determination.
Scored and classified outputs — triage severity, deterioration trajectory, referral matching, predictive maintenance. Logged separately from the human determination it informs.
No AI occupies the policy layer. Admissibility, consent validity, clinical determination, prescribing authority — every one is a human decision, backed by a structured, logged AI input. This separation is the legal and clinical foundation of the architecture, and it is enforced at every zone boundary rather than assumed.
This line holds in every Enkidu sector — see Trust.
Ninkarrak is Baru, tuned for care settings: the same platform, carrying clinically oriented sensor models, health-data handling posture, and the privacy obligations of patient information. What Baru guarantees everywhere — evidence behind every element, a human behind every decision — Ninkarrak guarantees at the point of care: a clinician decides, every time.
Every zone runs on the same shared stack — one layer for where things may happen, one for whether they may. Anshar gives each step explicit entry and exit conditions so nothing advances until its preconditions are met; Kittu governs what each agent may do and records the evidence behind every input a clinician sees. The full detail lives on the platform page.
Book a walkthrough of the zone architecture and see the clinical and operational continuums running end to end.