Healthcare AI· 6 min read
RAG for healthcare portals: useful answers without leaking patient data
Retrieval-augmented generation lets a portal answer from a patient’s own records. Here is how to do it so the answers are grounded, auditable and private.
Why plain chatbots fail in healthcare
A general model answers from what it learned in training. In a clinical setting that is the wrong source: the question is about this patient, this report, this hospital’s instructions.
Retrieval-augmented generation (RAG) changes the source. The system first finds the relevant passages — a discharge summary, a lab report, a preparation sheet — and only then asks the model to answer using those passages.
Four rules we build in
Retrieve only what the signed-in person may see. Access control is applied before retrieval, never after generation.
Show the source. Every answer links back to the document and date it came from, so a clinician can check it in one click.
Keep data in your tenancy. Embeddings, indexes and model calls stay inside your own cloud project and region.
Let it say “I don’t know”. When nothing relevant is retrieved, the assistant hands over to a person instead of guessing.
Where it pays off first
Report explanations in plain language, appointment preparation, discharge instructions and staff-facing “ask the record” search are the quickest wins. They are high-volume, low-risk and easy to measure.