STCC Triage Agent - DSPy Extension (v2.0.0)

Professional medical triage system powered by DSPy and DeepSeek, with 10 specialized nurse agents.

⚠️ IMPORTANT: Educational and research use only. NOT approved for clinical use.

---

Demo

Interactive triage system with 10 specialized nurse agents. Watch full video

---

What's New in v2.0.0

This is now a professional DSPy extension that users can install locally:

• 📦 PyPI package: pip install dspy-stcc-homecare
• 🚀 CLI commands: stcc-ui, stcc-optimize, stcc-api
• 🏗️ Clean structure: All code in stcc_triage/ package
• 📚 Bundled data: 225 STCC protocols pre-parsed and included
• 💾 Zero setup: Just install and run - no manual configuration
• 🔄 Breaking changes: Clean v2.0.0 release (no backward compatibility)

---

Installation

Quick Start (PyPI)

# Install the package
pip install dspy-stcc-homecare

# Configure API key
export DEEPSEEK_API_KEY="your-key-here"

# Launch UI
stcc-ui

Development Setup

# Clone repository
git clone https://github.com/chenhaodev/dspy-stcc-homecare.git
cd dspy-stcc-homecare

# Install in editable mode
pip install -e .

# Or use uv (faster)
uv pip install -e .

# Install with API support
pip install -e ".[api]"

Configure API Key

# Option 1: Environment variable
export DEEPSEEK_API_KEY="your-key-here"

# Option 2: .env file (for development)
cp .env.example .env
# Edit .env and add: DEEPSEEK_API_KEY=your_key_here

---

Quick Start

Launch Web UI

stcc-ui
# Opens at http://localhost:8501

Features:

• 🏥 Select from 10 specialized nurses
• 💬 Interactive chat interface
• 🎨 Color-coded triage levels
• 🔍 View reasoning steps
• 🔧 Check optimization status

Python API

from stcc_triage import STCCTriageAgent
from stcc_triage.nurses import WoundCareNurse

# Use baseline agent
agent = STCCTriageAgent()
result = agent.triage("55-year-old with severe chest pain and shortness of breath")

print(f"Triage Level: {result.triage_level}")
print(f"Justification: {result.clinical_justification}")

# Use specialized nurse (requires compilation first)
nurse = WoundCareNurse()
result = nurse.triage("deep laceration with bleeding")

---

Specialized Nurses

Instead of one generic agent, this system provides 10 specialized nurses, each optimized for their domain:

Nurse	Specialization	Protocol Count	Example Cases
Wound Care	Trauma, burns, bleeding	24	Deep laceration, severe burn
OB/Maternal	Pregnancy, labor	13	Contractions at 32 weeks, bleeding
Pediatric	Children/infants	12	Infant fever, child vomiting
Neuro	Stroke, seizure	7	Sudden weakness, worst headache
GI	Abdominal, digestive	6	Severe abdominal pain, GI bleeding
Respiratory	Breathing, asthma	6	COPD exacerbation, wheezing
Mental Health	Behavioral crises	5	Suicide risk, panic attack
CHF	Heart failure	4	Dyspnea, edema, chest pain
ED	Emergency/acute	All	Multi-trauma, critical
PreOp	Pre-surgical	2	Surgical clearance

Each nurse is optimized with domain-specific training data using DSPy's BootstrapFewShot.

---

Optimize Nurses

Optimize a Specific Nurse

stcc-optimize --role wound_care_nurse

# Other roles: ob_nurse, pediatric_nurse, neuro_nurse, gi_nurse,
# respiratory_nurse, mental_health_nurse, chf_nurse, ed_nurse, preop_nurse

Output:

Compiling Specialized Agent: Wound Care Nurse
Training set: 16 specialized cases
Distribution:
  emergency: 4 cases
  home_care: 4 cases
  moderate: 4 cases
  urgent: 4 cases

Optimizing for Wound Care Nurse...
This may take 5-10 minutes...

✓ Compiled Wound Care Nurse agent saved to:
  user_data/compiled/compiled_wound_care_nurse_agent.json

Optimize All Nurses

stcc-optimize

This compiles all 10 specialized nurses (takes ~1 hour).

---

Launch API Server

# Launch FastAPI server
stcc-api

# With auto-reload for development
stcc-api --reload

# Custom host/port
stcc-api --host 127.0.0.1 --port 8080

Visit http://localhost:8000/docs for interactive API documentation.

Example API Usage:

# Basic triage
curl -X POST "http://localhost:8000/triage" \
  -H "Content-Type: application/json" \
  -d '{"symptoms": "severe chest pain for 30 minutes"}'

# Specialized nurse triage
curl -X POST "http://localhost:8000/triage/specialized" \
  -H "Content-Type: application/json" \
  -d '{"symptoms": "deep laceration with active bleeding", "nurse_role": "wound_care_nurse"}'

---

Architecture

Package Structure

dspy-stcc-homecare/
├── stcc_triage/              # Main package
│   ├── core/                 # Core triage logic
│   │   ├── agent.py          # STCCTriageAgent
│   │   ├── signatures.py     # DSPy signatures
│   │   ├── settings.py       # DeepSeek config
│   │   └── paths.py          # Path management
│   │
│   ├── nurses/               # Specialized nurses
│   │   ├── roles.py          # NurseRole enum
│   │   └── specialized.py    # WoundCareNurse, OBNurse, etc.
│   │
│   ├── datasets/             # Dataset generation
│   │   ├── schema.py         # PatientCase schema
│   │   ├── generator.py      # Dataset generator
│   │   └── cases/            # Case definitions
│   │
│   ├── optimizers/           # Optimization logic
│   │   ├── metric.py         # Safety metrics
│   │   ├── optimizer.py      # BootstrapFewShot config
│   │   └── compiler.py       # Compilation logic
│   │
│   ├── protocols/            # Protocol handling
│   │   ├── parser.py         # Protocol parser
│   │   └── context.py        # Context enrichment
│   │
│   ├── ui/                   # Streamlit UI
│   │   ├── app.py            # Main app
│   │   └── components/       # UI components
│   │
│   ├── api/                  # FastAPI deployment
│   │   ├── app.py            # FastAPI app
│   │   └── models.py         # API models
│   │
│   ├── cli/                  # CLI commands
│   │   ├── ui.py             # stcc-ui
│   │   ├── optimize.py       # stcc-optimize
│   │   ├── api.py            # stcc-api
│   │   └── parse.py          # stcc-parse-protocols
│   │
│   └── data/                 # Bundled data
│       └── protocols/        # STCC protocols (~2MB)
│
├── user_data/                # User-generated (gitignored)
│   ├── compiled/             # Compiled nurses
│   └── datasets/             # Generated datasets
│
└── protocols/                # Generated protocols.json

How It Works

1. Protocol Context: Enriches patient symptoms with relevant STCC protocol guidelines
2. DSPy ChainOfThought: Structured reasoning with transparent decision-making
3. DeepSeek R1: Advanced reasoning engine with medical knowledge
4. BootstrapFewShot: Automatic optimization with domain-specific examples
5. Safety Metrics: Prevents under-triage (missing emergencies = score 0.0)

---

CLI Reference

# Launch Web UI
stcc-ui

# Optimize nurses
stcc-optimize                           # All nurses
stcc-optimize --role wound_care_nurse   # Specific nurse
stcc-optimize --regenerate-data         # Force regenerate training data

# Launch API server
stcc-api                                # Default: 0.0.0.0:8000
stcc-api --host 127.0.0.1 --port 8080   # Custom host/port
stcc-api --reload                       # Auto-reload for dev

# Parse protocols
stcc-parse-protocols                    # (Optional) Re-parse STCC markdown files

---

Python API Reference

# Core imports
from stcc_triage import STCCTriageAgent, TriageSignature, FollowUpSignature

# Specialized nurses
from stcc_triage.nurses import (
    WoundCareNurse,
    OBNurse,
    PediatricNurse,
    NeuroNurse,
    GINurse,
    RespiratoryNurse,
    MentalHealthNurse,
    CHFNurse,
    EDNurse,
    PreOpNurse,
    GeneralNurse,
    NurseRole,
)

# Optimization
from stcc_triage.optimizers import optimize_nurse, load_compiled_nurse

# Dataset generation
from stcc_triage.datasets import generate_all_specialized_datasets

# Protocol parsing
from stcc_triage.protocols import parse_all_protocols

---

Development

Run Tests

pytest tests/

Code Quality

# Format code
ruff format .

# Lint code
ruff check .

# Type checking
mypy stcc_triage/

---

Troubleshooting

"protocols.json not found"

This should only happen in development. The protocols are pre-bundled when installing from PyPI.

# For development only - regenerate protocols
stcc-parse-protocols

"Compiled agent not found"

stcc-optimize --role wound_care_nurse

Import errors after installation

pip install -e .  # Reinstall in editable mode

---

License

MIT License - See LICENSE

---

Contributing

Contributions welcome! This is an educational project demonstrating DSPy optimization for domain-specific agents.

---

Citation

If you use this project in research:

@software{stcc_triage_agent,
  title = {STCC Triage Agent: DSPy Extension for Medical Triage},
  author = {STCC Triage Agent Contributors},
  year = {2025},
  url = {https://github.com/chenhaodev/dspy-stcc-homecare}
}

---

Acknowledgments

• STCC Protocols: Schmitt-Thompson Clinical Content (225 protocols)
• DSPy: Stanford NLP's programming framework for LMs
• DeepSeek: Advanced reasoning engine with medical knowledge

---

Built with DSPy • Optimized for Safety • Educational Use Only