The system addresses a core operational challenge in emergency and inpatient care: accurate, transparent, and real-time patient prioritization.

• Machine Learning (XGBoost)
• Rule-based clinical safety overrides
• Explainable AI (SHAP)
• Local LLM-powered clinical assistance (Llama 3 via Ollama)

MedCognis delivers interpretable, data-driven risk scoring to support faster clinical decisions and optimized patient flow.

🧠 Hybrid Risk Scoring Engine

• XGBoost-based triage classification
• Hard-coded safety overrides (critical BP, SpO₂, HR thresholds)
• Multi-layer prioritization logic

The core of MedCognis is its dual-layer AI engine that combines the predictive power of XGBoost with the clinical rigor of Safety Overrides.

1. Structured Data Analysis: The engine processes 10+ clinical parameters (Age, BP, HR, SpO₂, etc.) to identify complex risk patterns.
2. Clinical Safety Overrides: Before final scoring, the system checks for critical threshold violations (e.g., Temp > 40°C or SpO₂ < 90%) to ensure high-risk patients are never underestimated by purely statistical models.
3. Real-time Explainability: Using SHAP, the engine identifies exactly which features (like high Heart Rate or Age) contributed most to the risk score, providing clinicians with immediate diagnostic context.

• Integrated SHAP (SHapley Additive exPlanations)
• Feature-level transparency for every prediction
• Clinician-readable reasoning outputs

• Symptom-vital correlation engine
• Condition prediction
• Specialist recommendation routing

• Glassmorphism-inspired UI
• ICU load tracking
• Department capacity visibility
• Vital trend radar & bar charts

• Heuristic PDF/Text ingestion
• Rapid structured data extraction
• In-memory preprocessing pipeline

• Secure, on-premise clinical assistant
• Powered by Llama 3 via Ollama
• No external data transmission

MedCognis follows a decoupled, service-oriented architecture optimized for low-latency clinical operations.

• Framework: Next.js (App Router)
• UI Styling: Tailwind CSS v4
• Data Visualization: Recharts
• Animations: Framer Motion
• Icons: Lucide React
• State Management: React hooks integrated with triage utilities

Orchestration between Frontend, ML models, and Database. Key Endpoints:

1. Patient registration
2. Risk prediction
3. Model retraining
4. Clinical AI chat

• Predictive Model: XGBoost classifier
• Data: Trained on structured clinical features
• Explainability: SHAP-based interpretability module

• Local LLM endpoint
• Ollama runtime
• Offline inference capability

• Transactional local database
• Patient records storage
• Role-Based Access Control (RBAC)
• Visit history tracking

• Node.js ≥ 18
• Python ≥ 3.10
• Ollama installed and running

cd Models
python -m venv .venv
source .venv/bin/activate     # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python app.py

# From project root
npm install
npm run dev

Ensure Ollama daemon is active:

ollama pull llama3

Default endpoint: http://localhost:11434

1️⃣ Access Dashboard Open: http://localhost:3000

2️⃣ Patient Triage

• Navigate to Patient Records
• Upload CSV / JSON / PDF
• Trigger AI risk assessment

3️⃣ Analytics View

• Inspect SHAP explanations
• Review risk breakdown
• Examine recommended specialists

4️⃣ Clinical Assistant

• Access Support & Help
• Interact with local LLM for structured guidance

├── app/                 # Next.js App Router
├── Models/              # Backend & ML Logic
│   ├── app.py
│   ├── triage_logic.py
│   └── triage_xgboost.pkl
├── components/          # Reusable UI Components
├── lib/                 # Frontend utilities
├── public/              # Static assets
├── types/               # TypeScript interfaces
└── README.md

• DICOM imaging integration (MRI/CT)
• Predictive bed capacity forecasting
• Federated learning model updates
• Multi-branch hospital synchronization
• Advanced anomaly detection in vitals

Team Name: Phoenixphones Clinical AI Lead: Phoenixphones teams

Members:

1. Naveen (Leader)
2. Prashant Gupta (Frontend Developer)
3. Sachin Chauhan (Synthesis Data Vault Generation and Research Work)
4. Bibek Kumar Sah (Resources Generator and Idea Creator)

Licensed under the MIT License. See LICENSE file for details.

MedCognis Health is a clinical decision support system. It is not a substitute for licensed medical judgment and must be used under professional supervision.