Apply for a GPU community grant: Academic project
Relational AI for Nursing: The First Open-Source LLM for Person-Centred Clinical Documentation
We are developing an open-source AI system to transform nursing documentation by embedding health equity and person-centred care principles directly into clinical AI.
The Problem
Current clinical AI tools generate generic, often dehumanizing documentation that fails to account for patient diversity—particularly in skin tone assessment for pressure ulcer risk, leading to documented health disparities.
Our Solution
Fine-tuned Llama-3 (8B) on 6,698 instruction pairs from the Foundation of Nursing Studies (FONS) literature
Equity Safety Gates: Mandatory skin tone documentation (Fitzpatrick/Monk scale) for wound assessments
Empathy Index: First-of-its-kind scoring system (1-5) for therapeutic engagement quality
FHIR Implementation Guide: Interoperable clinical data standards at opennursingcoreig.com
Evaluation Results
Equity (Skin Tone): 8/10 — Best-in-class performance
Person-Centredness: 7.6/10
Clinical Accuracy: 6.6/10
Open-Source Artifacts
🤗 Model: NurseCitizenDeveloper/nursing-llama-3-8b-fons
🤗 Space: relational-ai-4-nursing
📋 FHIR IG: opennursingcoreig.com
📂 GitHub: ClinyQAi/open-nursing-core-ig
Why We Need Support
GPU compute for extended training and inference hosting would help us:
Train additional epochs to improve clinical accuracy
Host the Gradio demo reliably for nurse testing
Expand the model to support more nursing specialties
License: CC BY-NC 3.0 (Non-Commercial, aligned with FONS source material)