A Voice-Enabled Palliative Care Assistant with Clinician Review and Preliminary Expert Evaluation

Palliative care guidance must account for symptoms, available resources, and the patient's goals of care. We present a working prototype that combines retrieval-augmented response generation with voice interfaces and a web application for clinician review. The demonstration traces a fictional caregiver query through a configured voice path, then opens the archived response for source inspection and structured review. Four palliative care physicians assessed the same 40 constructed pulmonary vignettes, yielding 160 reviews. The mean overall score was 7.80/10; 95/160 reviews recommended use without revisions, while 22/160 assigned safety ratings of 2 or below on a five-point scale. Absolute-agreement reliability was low, with single-rater ICC 0.13. The evaluated responses used vector retrieval, and the study does not validate the broader hybrid retrieval architecture, speech performance, or clinical outcomes. We describe the operational workflow, the evidence available for each component, and the limits that must precede further clinical use.