INDONESIA'S CLINICAL AI INFRASTRUCTURE · LIVE IN NTT

Indonesia is digitising
280 million patients.
Nobody owns the layer underneath.

Three government mandates now reach every layer of Indonesian healthcare — community, primary care, hospital. None of them came with a platform. SahAIbat operates across all three as one connected patient record, and trains Indonesia's own clinical language model on the consented data that record produces.

Investor brief →Enterprise

0.0M

community health workers

doctors

patients

One

connected record

ONE RECORD, MOVING THROUGH EVERY LAYER

CommunityPosyandu

MaternalBidan

FamilyWhatsApp

ClinicDoctor

HospitalClaim

PSE Kominfo· NIB 1202260248509

UU PDP· AES-256-GCM

SATUSEHAT· HL7 FHIR R4

AWS Jakarta· ap-southeast-3

SahAIbatsahabat · companion

THE NAME

Sahabat is the Indonesian word for companion — the friend who stays. We put AI in the middle of it, because that is where it belongs: inside the relationship, helping the person doing the work. Not in front of them, replacing it.

It is also the whole product decision. Every layer we build assists a human who keeps the final say — a kader, a midwife, a doctor, a coder. None of them are being automated away.

THE PLATFORM

Everyone in Indonesian healthtech builds one layer. The patient moves through all of them.

There are competent products at almost every layer of Indonesian healthcare — clinic EMRs, AI scribes, consumer telehealth apps, casemix consultancies. Every one of them is a point solution, and not one of them hands the next layer a record. That is the honest gap, and it is the whole thesis: SahAIbat runs the same system of record from the village health post to the hospital claim, which means each layer arrives at the next already knowing the patient.

LAYERMARKETSAHAIBAT

COMMUNITY

Kader

Kader · community health workers

Screening, growth tracking and referral at the Posyandu (village health post), aligned to the ILP primary-care standard.

WHAT THE MARKET OFFERS TODAY — PAPER REGISTERS

Paper registers. No commercial vendor finds this layer economic.

Nobody else is at every layer. That is not a marketing claim — it is a build order.

A single-layer competitor can add a second layer. What they cannot do is reconstruct the years of longitudinal, consented history that only exists because the platform was already in the field at the layer below.

THE ENGINE ROOM

The mandate asks for a form. We built the mathematics.

Digitising a register satisfies a regulation. It does not tell anyone whether a child is wasting or whether an outbreak has started. Underneath every interface we ship, a real model is running — which is why one measurement that costs a kader thirty seconds can end up as a district's early-warning signal without a single person re-entering it.

ONE MEASUREMENT. SIX LEVELS OF MEANING.

FIELD

Measured

A kader weighs and measures a child at the Posyandu. Thirty seconds, on a phone that may have no signal all day.

8.2 kg · 74 cm

14 months, female

ON DEVICE

Scored

WHO growth standards run on the handset — weight-for-age, height-for-age, weight-for-height — returned as Z-scores before the family stands up.

WAZ −2.7

WHO growth standard

RISK ENGINE

Classified

A WAZ of −2.7 is not a number a kader should have to interpret. It comes back as SAM, with the referral already written.

SAM

severe acute malnutrition

CLINICAL

Escalated

The midwife and the Puskesmas receive the case with the measurements attached — not a phone call describing them from memory.

Referral

midwife + Puskesmas notified

B2G

Aggregated

The same record updates village prevalence, district SAM rate, immunisation coverage and Posyandu performance ranking. Nobody re-types anything into a monthly report.

13% → 17%

district SAM prevalence

SURVEILLANCE

Watched

When communicable disease reports cross mean + 1.5 SD of that district's own history, the epidemic curve raises an SKDR-compatible alert on its own.

55 / week

alert threshold, auto-calculated

Under 30 seconds per child

Growth engine

WHO WAZ, HAZ and WHZ computed offline on the handset — classified, and referred, from the same screen.

Simple inputs → clinical meaning

Risk engine

Danger signs, ANC 10T completeness, weight velocity and immunisation gaps come back as graded risk, not raw rows.

SKDR-compatible

Surveillance engine

Alert thresholds computed as mean + 1.5 SD of a district's own history — not a national constant that fits nowhere.

Deterministic, not generative

Clinical engine

Labs, imaging and ECGs read together; eGFR and FIB-4 computed; the ICD-10 code held against the patient's own results.

Adaptive, inside WhatsApp

Assistant engine

The plan, the reminders and the language shift with each household's own history — not one template broadcast to everybody.

GOVERNMENT DASHBOARD · LIVE

The district sees what the village sees — the same day.

Posyandu ranking, nutrition status by WAZ band, immunisation coverage, stunting prevalence month by month and a live epidemic curve — generated from records a kader created that morning, with no reporting cycle in between.

WAZ

SAM · MAM · normal · over, banded automatically

10T

ANC completeness scored per pregnancy

SKDR

epidemic curve with auto-calculated thresholds

manual re-entry between field and government

EPIDEMIC CURVE — COMMUNICABLE DISEASE, ONE DISTRICT

Weekly casesAuto-calculated alert thresholdAbove threshold — SKDR signal

SOVEREIGN MODEL · IN TRAINING

Every layer above is also a training set.

The same consented records feed Indonesia's own clinical model. We are fine-tuning MedGemma on Indonesian clinical language — how a kader records a danger sign, how a midwife documents ANC 10T, how a doctor writes an assessment in Bahasa Indonesia, how a coder justifies a severity level under BPJS. Extraction runs today on our own GPU in Jakarta; nothing a doctor corrects leaves the country.

MedGemmamedical-specialist base, fine-tuned on Indonesian clinical text

🇮🇩 Jakarta GPUself-hosted inference — the machine is ours, not rented per call

Consent-boundcorrections become training data only where consent covers it

An app that satisfies a mandate can be rebuilt in a quarter. A risk engine a health ministry trusts, running on records traceable to the kader who took them, cannot — and that is the part this page is too short to do justice to.

See the technical brief →Talk to us

LIVE NOW · SAHAIBAT DOK

The clinic layer is already in doctors' hands.

DOK is the commercial engine of the platform and it is shipping today — an Indonesian clinical intelligence that reads the labs, X-rays and ECGs a patient brings, writes the note, checks the ICD-10 code against the patient's own results, and pre-checks the BPJS claim before it is submitted.

Reads lab panels, radiology, ECG and ultrasound together
Catalogue-bound ICD-10 — it cannot invent a code
BPJS kapitasi, Fornas and claim pre-flight built in
SATUSEHAT HL7 FHIR R4, submitted automatically

Visit sahaibatdok.com →Free for 30 days · no card

Konsultasi · Ny. Kartika S.

Kreatinin1.9 mg/dL ↑

Ureum58 mg/dL ↑

⚠ eGFR 40 — fungsi ginjal stadium G3b

Dihitung dari kreatinin ini. Tidak tercetak di laporan.

ICD-10E11.2dengan komplikasi ginjal

TRACTION

Running in the field. Not projected.

The platform is deployed today with real health workers, real clinicians and real partners — which is a different conversation from a roadmap.

children monitored in NTT

WHO

Z-scores computed on every visit

SAM / MAM

flagged and referred from the field

Growth tracked at the Posyandu against WHO standards — weight-for-age, height-for-age and weight-for-height — with malnutrition classified and referred from the same screen the measurement was taken on.

DEPLOYMENT PARTNERS

Yayasan Pijar TimurCommunity deployment partner · Nusa Tenggara TimurPAPHAPublic health associationPERDHAKINational faith-based hospital and clinic network

THE TEAM

Built by people who have shipped it.

A working platform across five products and three care layers, already in the field — built by a team small enough to still be moving quickly.

Sanjib MaityFounder · CEO & CTO

Draws the thing on a whiteboard, then writes the code that makes it true.

Dr. Ratih Rakhmawati, M.BiomedClinical validation

Decides whether a model's answer would survive a real consultation — and sends it back until it would.

Surabhi DasClinical & medical concept

Turns a national protocol into something a screen can actually ask, in the order a clinician asks it.

Stefanus BereField & partnerships · NTT

The reason a Posyandu in Timor trusts software it never asked for.

Shindy FarahOperations · Indonesia

Holds the gap between a plan written in Jakarta and a health post that has to run it on a Friday.

Saurav DasInfrastructure & DevOps

Owns the servers, the GPUs and the 3am pager. Nothing ships until it stays up.

Chief Market StrategistOpen role

Owns how Indonesia hears about all of this.

Business DevelopmentOpen role

Turns a working platform into signed clinics.

Inside SalesOpen role

First voice a doctor hears after a trial begins.

Live in Timor Tengah Utara, Nusa Tenggara Timur.

If you see what we see, we should talk.

This page makes the argument. The numbers, the structure and the timeline are in the deck and in the conversation — not published here.

See the deck →Talk to us

Viantra Health · 11679210 Canada Inc.

Indonesia is digitising280 million patients.Nobody owns the layer underneath.