Every note Asternoos produces is a draft. AI-generated clinical documentation is intended to assist clinicians and must be reviewed, corrected, and approved by a licensed healthcare professional before it is relied upon for patient care or entered into a medical record.
1. What the AI does
Asternoos uses two kinds of AI model:
- Speech recognition converts consultation audio into a text transcript.
- Large language models read that transcript and draft a structured clinical note — for example history, examination, assessment, and plan — in the format you have configured.
We also use a vision-capable model to read uploaded lab reports and images and extract structured fields into a patient record.
The AI reorganises and summarises what was said. It is not a reasoning engine for clinical decisions and has no access to guidelines, drug interaction databases, or the patient's full medical history unless you provide it.
2. What the AI does not do
- It does not diagnose, prescribe, triage, or recommend treatment.
- It does not verify anything it writes against a clinical knowledge base.
- It does not know when it is wrong, and it will not warn you.
- It does not take any action on your behalf. Nothing is sent, filed, or submitted anywhere without you doing it.
3. Not a medical device
Asternoos has not been assessed or certified under the EU Medical Device Regulation, the Turkish Medical Device Regulation, the US FDA framework, or any equivalent regime. It carries no CE mark as a medical device and is not intended to diagnose, treat, cure, mitigate, or prevent disease. It is a documentation tool.
4. Known limitations
These are real failure modes, observed with this class of technology. Please read them.
- Hallucination. Language models can produce fluent, confident text describing findings, symptoms, or history that were never mentioned in the consultation.
- Omission. Clinically important detail that was said may be left out of the summary, particularly when it was mentioned briefly or in passing.
- Numbers, doses, and units. Drug names, dosages, frequencies, laboratory values, and units are the highest-risk items to mis-transcribe. Verify each one against the source.
- Negation. "No chest pain" can be transcribed or summarised as "chest pain". Check that negatives are preserved.
- Audio conditions. Accents, dialects, fast or overlapping speech, background noise, and poor microphone placement all degrade transcription accuracy.
- Multiple speakers. Statements may be attributed to the wrong speaker, including confusing what the patient reported with what the clinician concluded.
- Language mixing. Turkish consultations containing English or Latin medical terminology can be transcribed inconsistently.
- Wrong patient. If a recording is started on the wrong patient record, the AI will not detect this. Confirm the patient before recording and before saving.
- Non-determinism. The same audio can produce slightly different notes on different runs.
5. Your responsibility as a clinician
By using Asternoos you accept that you are the human in the loop and the clinical author of the final note. Specifically, you must:
- Read the entire note before saving, sharing, signing, or entering it in a medical record.
- Verify every medication, dose, number, laboratory value, allergy, and negative finding against your own recollection and the source material.
- Correct anything inaccurate, and delete anything that was not actually discussed.
- Confirm the note is attached to the correct patient.
- Apply your own clinical judgement. The note reflects what the model extracted, not what is clinically correct.
Asternoos does not replace your professional judgement, your record-keeping obligations, or your duty of care.
6. Do not use in emergencies
Asternoos is not suitable for emergency, time-critical, or life-support situations. Do not delay clinical action to wait for a generated note. Do not rely on the platform as your only means of producing required documentation; network outages, provider incidents, and processing failures do occur.
7. Transparency about the models we use
We use third-party foundation models from OpenAI and Google, accessed via their APIs. We configure and prompt these models; we do not build or train them. Model versions change over time as providers update them, which can change output style and behaviour. Your audio, transcripts, and notes are not used to train these models. See Subprocessors and Security.
8. Patients
If you are a patient, your consultation may be recorded and processed by AI only after your clinician has informed you and obtained your explicit consent. You can refuse, and refusing will not affect the medical care you receive. See the Patient Information Notice.
9. Reporting an AI error
If Asternoos produces a dangerous, fabricated, or seriously incorrect output, please tell us. Reports directly shape how we tune the system. Email hello@asternoos.com with what happened and, if you can, the note identifier. Do not include patient-identifying information in the email.