AI Ethics & Transparency
JegsMediScan uses a large language model as part of its symptom-check assessment. Here is exactly what that means, and what it doesn't.
Decision support, never autonomous diagnosis
The AI assessment suggests a care level and shows its reasoning — it does not diagnose, and it never finalizes a patient's acuity or disposition on its own. In the staff console, a nurse or physician can override any AI-suggested acuity at any time; overrides are logged, never discouraged. In the patient-facing symptom check, red-flag emergency symptoms are checked before anything else, and the tool tells a patient to call 911 rather than proceeding if they're present.
Built on published clinical reasoning, not invented rules
The model is instructed to apply widely used, published clinical decision tools (HEART, Wells, PECARN, and others) rather than reasoning from scratch — see our evidence page for the full list and citations. Our own risk and admission-prediction models are separate from this assessment flow entirely: they run in shadow mode only, evaluated against real triage outcomes for internal review, with no ability to influence a real patient's care today.
Third-party AI vendor, used under a governed, human-in-the-loop model
We rely on a third-party large language model provider rather than a model we trained ourselves. Any vendor we use for a clinical-adjacent use case is selected and governed under the same principle this whole page describes: the model supports a clinician, it never replaces one, and that constraint is enforced in how the product is built, not left to the model's own judgment.
What hasn't been validated yet
We don't claim clinical-grade accuracy for this product's AI output, and we removed that language from our marketing copy where it previously appeared. No independent clinical validation study has been run. See security & compliance for where our regulatory approach stands today.