SYSTEM_CARD.md

metadata

title: 'Building the PNA Assistant: A Nurse''s Guide to AI Resilience Tools'
authors:
  - user: NurseCitizenDeveloper
    guest: true
tags:
  - nlp
  - healthcare
  - gemma
  - rag
  - ethical-ai
  - system-card

Building the PNA Assistant: A Nurse's Guide to AI Resilience Tools

"AI isn't here to replace nurses. It's here to be built by us."

The resilience of the nursing workforce is a global priority. In the UK, the Professional Nurse Advocate (PNA) role has been a game-changer, using the A-EQUIP model to provide restorative clinical supervision. But with staffing pressures, accessing a trained PNA when you need one most can be a challenge.

What if we could build a tool to bridge that gap? Not to replace the human supervisor, but to provide a safe, 24/7 space for reflection?

Enter the PNA Assistant—an open-source AI research tool built by a nurse, for nurses.

🏥 What is the PNA Assistant?

The PNA Assistant is a specialized AI agent designed to simulate the "Restorative Clinical Supervision" function of a Professional Nurse Advocate.

Unlike generic chatbots, it doesn't just "chat." It is grounded in the A-EQUIP Model (Advocating for Educating for Quality Improvement and Practice) and strictly adheres to the official NHS PNA Handbook.

Key Capabilities

• Restorative Dialogue: Asking reflective questions to help nurses process emotional challenges.
• Policy Grounding: Retrieving exact guidance from the A-EQUIP framework (Normative, Formative, and Restorative).
• Safety First: Hard-coded safety rails to distinguish educational support from clinical/crisis intervention.

---

⚙️ Under the Hood: The "Nurse Citizen Developer" Stack

We built this tool using a Retrieval-Augmented Generation (RAG) architecture, hosted entirely on Hugging Face Spaces using ZeroGPU.

1. The Brain: Gemma 2 2B IT 🧠

We chose Google's Gemma 2 2B IT (Instruction Tuned) as our foundation. Why?

• Efficient: It runs fast on renewable cloud resources.
• Capable: It has excellent reasoning capabilities for its size.
• Open: It aligns with our philosophy of open healthcare research.

2. The Knowledge: RAG 📚

To prevent hallucinations (AI making things up), we don't rely solely on the model's training data. Instead, we use RAG:

1. Knowledge Base: We embedded the PNA Handbook and A-EQUIP Guidance.
2. Retrieval: When you ask a question, the system searches these documents for the answer.
3. Generation: The AI uses that official text to construct its response.

3. The Persona: "Compassionate Instruction Tuning" ❤️

The "secret sauce" is the System Prompt. We didn't just tell it to be helpful; we told it to be a Nurse.

• Legacy Challenge: AI can sound robotic or overly solution-focused.
• Our Solution: We instructed the model to prioritize Validation and Reflection over immediate problem-solving, mirroring the core skills of a PNA.

---

🚀 Why "Citizen Development" Matters

This project proves that you don't need a Computer Science degree to build impactful health AI. You need clinical expertise and curiosity.

By taking ownership of the technology, we ensured that Nursing Values—compassion, safety, and advocacy—were baked into the code from Day 1. This isn't just an AI tool; it's a statement that nurses are ready to lead in the digital era.

🎓 Research & Availability

The PNA Assistant is available today as a Research Artifact. We invite nursing researchers, educators, and digital health innovators to test it, break it, and build upon it.

• Try the Model: Hugging Face Space
• Read the Code: GitHub Repository

Disclaimer: This tool is for educational and research simulation purposes only. It does not provide clinical diagnosis or crisis mental health support.