Patient Input (Symptoms + Vitals)
    ↓
[1] Intake Agent → Parses free-text symptoms into structured data
    ↓
[2] Risk Prediction Agent → ML model predicts LOW/MEDIUM/HIGH (78% accuracy)
    ↓
[3] XAI Explanation Agent → SHAP explains why (which factors mattered?)
    ↓
[4] Treatment Agent → Generates evidence-based care recommendations
    ↓
[5] Drug Safety Agent → Checks for dangerous drug interactions
    ↓
[6] FHIR Formatter Agent → Produces HL7 FHIR R4 bundles + SOAP notes
    ↓
Final Output: Risk + Explanation + Treatment Plan + Safety Alerts + Clinical Documents

scikit-learn==1.3.0          # Machine Learning algorithms
numpy==1.24.3               # Numerical computations
pandas==2.0.3               # Data manipulation
shap==0.42.3                # Explainability
fastapi==0.104.1            # Web framework for APIs
uvicorn==0.24.0             # ASGI server
joblib==1.3.0               # Model serialization
requests==2.31.0            # HTTP requests
python-dotenv==1.0.0        # Environment variables
pytest==7.4.0               # Testing framework

clinexa-ai/
├── data/
│   ├── generate_data.py              # Generate 1000 synthetic patients
│   └── synthetic_triage.csv           # Output: synthetic patient data
│
├── phase1_ml/
│   ├── train_model.py                # Train RandomForest ML model
│   └── models/
│       ├── risk_model.joblib
│       ├── shap_explainer.joblib
│       ├── label_encoders.joblib
│       └── metrics.json
│
├── phase2_mcp/
│   ├── fhir_mcp/fhir_server.py      # FHIR Server (Port 8001)
│   ├── drug_mcp/drug_server.py      # Drug Safety Server (Port 8002)
│   ├── xai_mcp/xai_server.py        # XAI Server (Port 8003)
│   └── grok_mcp/grok_server.py      # Grok LLM Server (Port 8004)
│
├── phase3_agents/
│   └── orchestrator.py               # 6-agent pipeline orchestrator
│
├── tests/
│   └── integration_test.py           # End-to-end testing
│
├── .env                              # Environment variables
├── README.md                         # This file
├── requirements.txt                  # Python dependencies
└── LICENSE

Processing 3 synthetic patients end-to-end:

Patient 1 (SYN-10001):
  ✓ Risk prediction: HIGH (73% confidence)
  ✓ SHAP explanation: Generated
  ✓ Drug alert: Warfarin + Aspirin (HIGH risk) — Bleeding detected
  ✓ FHIR bundle: Created
  ✓ SOAP note: Generated

Patient 2 (SYN-10002):
  ✓ Risk prediction: MEDIUM (46% confidence)
  ✓ FHIR bundle: Created
  ✓ SOAP note: Generated

Patient 3 (SYN-10003):
  ✓ Risk prediction: HIGH (55% confidence)
  ✓ FHIR bundle: Created
  ✓ SOAP note: Generated

Overall: 100% SUCCESS — All systems operational

GROK_API_KEY=your-xai-api-key-here
FLASK_ENV=production
LOG_LEVEL=INFO

@software{clinexa_ai_2026,
  title={Clinexa AI: Multi-Agent Clinical Triage System},
  author={Kaleem, Laraib},
  year={2026},
  url={https://github.com/LaraibKaleem/clinexa-ai},
  version={1.0.0}
}

---

## Acknowledgments

- **Scikit-learn** — Machine Learning algorithms
- **SHAP** — Explainability framework
- **FastAPI** — Web framework
- **Railway** — Cloud deployment
- **OpenFDA** — Drug interaction database
- **HL7 FHIR** — Healthcare data standards
- **Prompt Opinion** — AI marketplace
- Open-source community

---

## Disclaimer

**This is a research and educational project.** It is designed for demonstration and learning purposes using 100% synthetic data. It is NOT intended for use in actual clinical settings without proper regulatory approval (FDA clearance, clinical validation, etc.).

Clinexa AI demonstrates proof-of-concept for multi-agent healthcare AI but should NOT be used to make real clinical decisions without qualified medical professionals and appropriate regulatory oversight.

---

Built with ❤️ for healthcare innovation
Last updated: May 2026
<!-- # Clinexa AI — Multi-Agent Clinical Triage System

## Elevator Pitch
6 AI agents that work like a hospital team —
from patient intake to FHIR output — in seconds.

---

## What It Does

A patient describes their symptoms.
Clinexa AI sends them through 6 agents:

1. Intake Agent — reads symptoms, structures them
2. Risk Agent — predicts LOW / MEDIUM / HIGH risk
3. XAI Agent — explains WHY using SHAP
4. Treatment Agent — recommends care plan
5. Drug Safety Agent — checks medication conflicts
6. FHIR Agent — produces HL7 R4 clinical report

---

## Tech Stack

- Python 3.11
- FastAPI — MCP servers
- Scikit-learn — RandomForest ML model
- SHAP — Explainable AI
- xAI Grok — LLM for clinical text
- HL7 FHIR R4 — clinical output standard
- MCP + A2A — agent communication
- Prompt Opinion — deployment platform

---

## Project Structure
clinexai/
├── requirements.txt
├── README.md
├── .env
├── data/
│   ├── generate_data.py
│   └── synthetic_triage.csv
├── phase1_ml/
│   ├── train_model.py
│   └── models/
├── phase2_mcp/
│   ├── fhir_mcp/fhir_server.py
│   ├── drug_mcp/drug_server.py
│   ├── xai_mcp/xai_server.py
│   └── grok_mcp/grok_server.py
├── phase3_agents/
│   └── orchestrator.py
├── phase4_deploy/
│   └── prompt_opinion_config.py
├── phase5_demo/
│   └── demo_script.txt
└── tests/
└── integration_test.py
---

## How To Run

### Step 1 — Install packages
```bash
pip install -r requirements.txt