AI Elements

The healthcare and life science parts of an AI interface, built on shadcn's components and its AI Elements. Use shadcn's for the chat, reasoning and tool calls, and these for the clinical record around them.

Draft the ED note for bay 4 and list the differential.

AI-generated draft

Not yet reviewed by a clinician. Check it against the record before you rely on it; it is not a diagnosis.

Example model · generated Oct 21, 2026, 9:30 AM UTC

Subjective

58-year-old with central chest tightness for two hours, worse on exertion. No previous cardiac history (no source).

Objective

Sinus rhythm at 78 bpm without ST changes; high-sensitivity troponin 18 then 21 ng/L, below the 99th percentile.

Assessment

Low risk of acute myocardial infarction on the 0/3-hour pathway; stable angina remains possible.

Plan

Outpatient exercise testing. Return if pain comes on at rest (no source).

Dotted text has no source. Check it against the record.

Stable anginaI20.8

42%

Gastro-oesophageal refluxK21.9

27%

CostochondritisM94.0

14%

Acute coronary syndromeI24.9

Model confidence, not a probability of disease. A clinician decides the diagnosis.

Emergency department note

Awaiting review. Review as Dr A. Rivera before this is used.

Patterns

Principles

Say what made it: AI disclaimer names the model and time, and changes once a clinician has reviewed the content.
Show the sources: AI note cites a source for each claim and marks the claims without one, so review starts where it's needed.
Keep confidence honest: Differential list shows the model's confidence as confidence, never as a probability of disease.
A clinician decides: Clinician review records who approved, asked for changes or rejected, when, and why.

These patterns help a clinician review AI output; they don't make it safe. Whether and how you use AI in care is a clinical, regulatory and organisational decision.