Research
Tuned on
real notes.
KnatHealth runs on a clinical model you can swap for your own fine-tune. Point the assessment engine at your model id and the documentation pipeline stays identical.
01
Fine-tuning corpus
Trained on de-identified consultation transcripts paired with physician-signed notes, so the model learns the house style of real documentation rather than textbook prose.
02
Calibrated confidence
Differential likelihoods are calibrated against physician-adjudicated outcomes. A 70% output should be right roughly seven times in ten — not merely confident-sounding.
03
Triage safety first
Red-flag recall is optimised ahead of precision. The model is tuned to over-escalate rather than miss an emergency presentation.
87%
Primary-care accuracy
99.1%
Red-flag recall
6 min
Median to draft plan
100%
Physician-signed plans
© 2025 KnatHealth · Clinical AI, humanly signedPrivacySecurityTerms