NEST: Newborn & Maternal Safe Transition

Link - https://www.youtube.com/watch?v=XNFU0ZbqBhs&t=3s

NEST is an advanced A2A (Agent-to-Agent) clinical orchestrator built for the Prompt Opinion platform (Agents Assemble Hackathon 2026). It solves one of the most dangerous gaps in healthcare: the fragmented postpartum handoff between maternal and infant charts.

Instead of a single LLM trying to guess guidelines, NEST binds the mother-infant dyad and convenes five specialized AI agents in parallel to generate a unified, evidence-backed discharge plan.

---

✨ Key Features

• Multi-Agent Council: Runs 5 parallel specialists (Maternal OB, Pediatrics, Lactation, Mental Health, Social Work) using the Google ADK.
• FHIR Dyad Binding: Automatically extracts both maternal and infant clinical data from a single maternal FHIR patientId provided by Prompt Opinion.
• Wearables Telemetry Stream: Integrates real-time data from Apple Watch (Maternal BP/HR) and Owlet Dream Sock (Infant SpO2/HR).
• Claude Code TUI Aesthetic: Returns a beautiful, high-signal Terminal UI (TUI) output with ASCII boxes, Kanban task boards, and color-coded alerts.
• Evidence Anchors: Combats hallucination by strictly linking every generated recommendation to curated clinical guidelines (ACOG, AAP, LactMed).

---

🧠 The Architecture

graph TD
    User([Clinician]) -->|Prompt Opinion UI| PO[Prompt Opinion Coordinator]
    PO -->|A2A Protocol + FHIR Context| NEST[NEST Orchestrator]
    
    NEST -->|Extracts Dyad| FHIR[(FHIR Server)]
    NEST -->|Reads Telemetry| Wearables[(Wearables API)]
    
    NEST --> OB[🩺 Maternal OB Agent]
    NEST --> Peds[👶 Pediatric Agent]
    NEST --> Lact[🤱 Lactation Agent]
    NEST --> MH[🧠 Mental Health Agent]
    NEST --> SW[🏠 Social Work Agent]
    
    OB --> Synthesize
    Peds --> Synthesize
    Lact --> Synthesize
    MH --> Synthesize
    SW --> Synthesize
    
    Synthesize[Unified TUI Report] -->|A2A Response| PO

Loading

---

🚀 Quick Start (Local Development)

1. Install Dependencies

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Configure Environment

Create a .env file in the root directory:

# Your Gemini / LiteLLM API Key
GOOGLE_API_KEY=your_gemini_api_key

# The API Key you will put into Prompt Opinion's Agent Card
API_KEYS=your_secret_a2a_key

# Model Selection
NEST_ORCHESTRATOR_MODEL=gemini/gemini-2.5-flash

3. Run the Server

uvicorn nest_agent.app:a2a_app --host 0.0.0.0 --port 8005

4. Expose via ngrok

ngrok http 8005

Take the resulting https://...ngrok-free.dev URL and append /.well-known/agent-card.json to use in Prompt Opinion.

---

☁️ Cloud Deployment (Render)

This repository is pre-configured for 1-click deployment on Render.com.

1. Connect this GitHub repository to Render as a Web Service.
2. Build Command: pip install -r requirements.txt
3. Start Command: uvicorn nest_agent.app:a2a_app --host 0.0.0.0 --port $PORT
4. Add your .env variables in the Render dashboard.

---

🛠️ Usage in Prompt Opinion

1. Go to Prompt Opinion -> External Agents -> Add Connection.
2. Paste your Agent Card URL: https://<your-domain>/.well-known/agent-card.json
3. Enter your API_KEYS value in the auth header.
4. Upload the Skill Packages located in nest_agent/prompt_opinion/po_skill_packages/.
5. Ask your coordinator: "Is the patient ready for discharge?" or "Check the wearable vitals for the mom and baby."

---

⚖️ License

MIT License. See LICENSE for details.