Virtual Patients Let Medical Students Practice Doctor-Patient Talks

A research team from Universidad de los Andes in Bogotá, with a collaborator at the University of Luxembourg, has developed a large language-model “virtual patient” for medical-student training that can simulate realistic doctor-patient interviews and grade performance automatically. Described in Discover Artificial Intelligence, the system targets anamnesis—structured history taking that can provide up to 80% of diagnostic information. Students pick from five adult cases, including type 2 diabetes, COPD, and hypothyroidism, and start without any chief-complaint prompt. The agent supports typed questions or browser speech input, replies with text and synthesized speech, and uses a directed-graph architecture plus semantic memory in PostgreSQL/pgvector. It also generates audio using five personality styles and grades using criteria derived from validated communication tools.