An AI scribe that speaks your clinic's language.
JONAH AI listens to the consultation in Bahasa Indonesia, writes the SOAP note the way a klinik pratama does, and can research a cited second opinion while you see the patient. It drafts. You decide. You sign.
What happens in this video
- Transcription in Bahasa Indonesia; English is detected and routed; other languages are marked, not guessed. Audio is never stored.
- Findings surface as chips; denied symptoms and other people's illness are struck out.
- Names, NIK and addresses are replaced before any text reaches the AI.
- A SOAP draft appears; the doctor inserts it field by field, and each insert is logged.
- A cited differential appears with sources for and against and amber danger signs; the doctor chooses.
From the conversation to a signed note, in five steps
Each step is visible to the doctor, and only the last one, signing, changes the medical record.
Bahasa first. English understood.
Indonesian clinics switch languages mid-sentence. JONAH routes each stretch of speech to the model that handles it best, so an English question does not come out as Indonesian nonsense.
- Transcript as flowing paragraphs, with speaker turns, timestamps and the raw text one tap away.
- Unknown languages are marked as [unknown speech] rather than invented.
- Fast on ordinary hardware. A 45-second consult transcribes in about a second, without a GPU.
TRANSKRIP
PASIEN Sudah tiga hari demam, Dok. Mual juga.
DOKTER Any travel outside Bali recently? EN
PASIEN Tidak, Dok. Di rumah saja.
KELUARGA [unknown speech] tidak ditebak
It knows "tidak" from "ya".
A scribe that writes "cough" because the patient said "no cough" is worse than no scribe. JONAH's finding layer strikes out denied symptoms and illness that belongs to someone else.
- Negation: "tidak batuk" becomes a struck-out batuk.
- Other people: "tetangga saya kena demam berdarah" is the neighbour's, not the patient's.
- Explained: hover any chip to see the words that triggered it.
A SOAP draft in seconds, in your format.
The draft follows the klinik-pratama layout doctors already write: KU and RPS, denials, vitals and general status, working diagnosis with its basis, then treatment, education and follow-up.
- No basis, no line. Anything the conversation did not support is left out, not filled in.
- Insert field by field. "Sisipkan ke catatan" puts one section into the note; edit it like your own.
- Vitals never come from speech. They come from triage, where they were measured.
RINGKASAN SOAP · DRAF
A second opinion that shows its work.
On request, JONAH researches the presentation in the background and returns a short differential: the top candidates, the evidence for and against each, what would tell them apart, and danger signs.
- Cited or not shown. A candidate without a source does not appear.
- Danger signs in amber, never red. Red in JONAH means an error, not an opinion.
- Only valid codes. "Terima sebagai diagnosis" is disabled when the ICD-10 code is not in the national dictionary.
- Runs in the background for one to three minutes while you keep examining.
Mendukung: demam 3 hari, ruam · 3 sumber
Pembeda: pola ruam, trombosit · 2 sumber
The guardrails
What JONAH AI does, and what it is built never to do.
It does
- Transcribe, summarise and suggest, in violet, clearly marked as AI.
- Work only on de-identified text.
- Log every insert as AI-accepted, by name.
- Step aside gracefully: if AI is down, the EMR keeps saving, signing and dispensing.
It never
- Writes to the medical record without the doctor's click.
- Stores consultation audio.
- Fills in vital signs from speech.
- Makes the diagnosis. It offers candidates; the doctor chooses.
Hear it on your own consult.
We will run the scribe live in a walkthrough, in Bahasa Indonesia, with your doctors.