Lead PI:
No lead investigators associated with this post.
Center Researchers:
Collaborating Partners
- Duke Health Nursing
- Weijun Li, Co-Lab / MPS Student Developer
- Iara Ravagni, Co-Lab / MPS Student Developer
The Challenge
Nurses encounter verbal and physical aggression more than any other clinical staff, yet training opportunities are limited. Standardized patient actors are costly and can’t scale to the dozens of practice repetitions learners need, leaving graduates under‑prepared for real-world conflict.
We wanted to use virtual reality to help nurses better prepare for these encounters.
Our Solution
We created a VR simulation that lets student and practicing nurses rehearse challenging bedside conversations with an aggressive patient. We developed the simulation in Unity for head-mounted displays, pairing interactive conversations with high-fidelity visuals so learners feel present at the bedside.
A large‑language model (LLM), such as ChatGPT, powers the avatar’s dialogue and emotions, giving learners a fresh, unscripted encounter every time. The LLM drives the virtual patient, modulating voice, body language, and emotional intensity in real time as the nurse attempts to calm the situation. Additionally, we developed the ability to host the LLM on a Duke server, which reduces latency. The server keeps all protected health information inside Duke, ensuring HIPAA compliance.
After each encounter, an AI coach reviews the transcript and automatically generates a PDF report that highlights strengths, missed opportunities, and recommended phrasing for future attempts.
The Duke Health Nursing plans to expand the program by working with an external development company.