Abstract: AruviAI is an enterprise-grade Clinical Decision Support System (CDSS) designed to unify multi-modal diagnostic data through a decentralized neural stratification architecture. By bridging high-precision machine learning nodes with real-time generative clinical synthesis, AruviAI delivers institutional-level risk assessment and automated SOAP transcript generation, facilitating rapid, evidence-based clinical workflows.
AruviAI operates on a Tri-Layer Stratification Architecture, ensuring a seamless flow from clinical data acquisition to diagnostic synthesis.
graph TD
%% Styling
classDef frontend fill:#61DAFB,stroke:#000,stroke-width:2px,color:#000
classDef backend fill:#009688,stroke:#000,stroke-width:2px,color:#fff
classDef ai fill:#4f46e5,stroke:#000,stroke-width:2px,color:#fff
classDef database fill:#336791,stroke:#000,stroke-width:2px,color:#fff
classDef external fill:#FF9900,stroke:#000,stroke-width:2px,color:#000
%% Client Layer
subgraph Client_Layer ["🖥️ Client Interfaces (React/Vite)"]
UI_Pat["Patient Dashboard"]:::frontend
UI_Doc["Clinician Workspace"]:::frontend
UI_Admin["Institutional Dashboard"]:::frontend
UI_Chat["Kira A.I. Chat Interface"]:::frontend
UI_3D["3D Anatomy Viewer"]:::frontend
end
%% API Gateway & Backend Layer
subgraph API_Gateway ["⚙️ Backend Layer (FastAPI)"]
API_Auth["Auth & JWT Router"]:::backend
API_Doc["Document & PDF Router"]:::backend
API_Ana["Analytics & Alerts Router"]:::backend
API_Consult["Human Consultations & WebSockets"]:::backend
end
%% Neural Orchestration Layer
subgraph Orchestration ["🧠 Neural Orchestration (src/coordinator)"]
Coord_Exec["Executor Engine"]:::ai
Coord_Cross["Cross-Intelligence Engine"]:::ai
Coord_Explain["Explainability Engine (SHAP)"]:::ai
Coord_Agg["Synthesis Aggregator"]:::ai
end
%% Specialized ML Nodes
subgraph ML_Nodes ["🔬 Specialized AI Agents (src/agents)"]
Agent_Kira["Kira Conversational Agent"]:::ai
Agent_CV["Brain Tumor CV Agent"]:::ai
Agent_Heart["Cardiovascular Agent"]:::ai
Agent_Metabolic["Metabolic/Diabetes Agent"]:::ai
Agent_Renal["Renal/Kidney Agent"]:::ai
Agent_Hepatic["Hepatic/Liver Agent"]:::ai
end
%% Persistence & External Services
subgraph Data_Services ["💾 Persistence & External APIs"]
DB[(PostgreSQL Store)]:::database
Cache[(Memory Lattice)]:::database
LLM_Gemini["Google Gemini API"]:::external
LLM_Groq["Groq (Llama 3)"]:::external
end
%% Connections
UI_Pat -->|REST/WS| API_Gateway
UI_Doc -->|REST/WS| API_Gateway
UI_Admin -->|REST/WS| API_Gateway
UI_Chat -->|WebSocket| API_Gateway
UI_3D -->|REST| API_Doc
API_Auth -.->|Auth/Read/Write| DB
API_Doc -.->|Read/Write| DB
API_Consult -.->|State| Cache
API_Gateway -->|Routing| Orchestration
Coord_Exec --> Agent_Kira
Coord_Exec --> Agent_CV
Coord_Exec --> Agent_Heart
Coord_Exec --> Agent_Metabolic
Coord_Exec --> Agent_Renal
Coord_Exec --> Agent_Hepatic
Agent_Kira -->|NLP| LLM_Groq
Agent_Kira -->|Vision| LLM_Gemini
Agent_CV --> Coord_Explain
Agent_Heart --> Coord_Explain
Agent_Metabolic --> Coord_Explain
Agent_Renal --> Coord_Explain
Agent_Hepatic --> Coord_Explain
Coord_Explain --> Coord_Agg
Coord_Agg -->|SOAP/PDF Gen| API_Gateway
sequenceDiagram
autonumber
actor Patient
participant Interface as Kira A.I. / Web UI
participant Backend as FastAPI Gateway
participant Cord as Neural Orchestrator
participant ML as Specialized AI Nodes
participant LLM as Generative Synthesis (LLM)
participant DB as PostgreSQL DB
actor Clinician
%% Step 1: Data Intake
Patient->>Interface: Inputs symptoms / Uploads Medical PDF
Interface->>Backend: Secure Payload (JWT Auth)
%% Step 2: Routing & Parsing
Backend->>Cord: Initialize Assessment Request
Cord->>DB: Fetch historical patient state
%% Step 3: Stratification
par Feature Extraction
Cord->>ML: Image Payload -> CV Node (EfficientNet)
and Structured Data Inference
Cord->>ML: Vital Signs -> XGBoost/RF Nodes
end
%% Step 4: Inference & XAI
ML-->>Cord: Raw Risk Probabilities (e.g. 0.89 Risk)
Cord->>Cord: Explainability Engine runs SHAP values
%% Step 5: Generative Synthesis
Cord->>LLM: Send structured data + risk factors
Note over LLM: Models clinical guidelines<br/>(ADA, AHA, etc.)
LLM-->>Cord: Generates SOAP Transcript & Recommendations
%% Step 6: Alerting & Storage
Cord->>Backend: Return Aggregated Synthesis
Backend->>DB: Save Consult & Transcript
%% Step 7: Clinician Alerting
alt Risk > Threshold
Backend-->>Clinician: WebSocket 🚨 CRITICAL_RISK_ALERT
else Routine Check
Backend-->>Clinician: Queue Update via Dashboard
end
%% Step 8: Review & Consult
Clinician->>Interface: Reviews SOAP Transcript & SHAP plots
Clinician->>Patient: Initiates Telemedicine Video Consult
- Tri-Role Access Control: Granular permissioning models for Patients (Historical telemetry, Alerts), Clinicians (Diagnostic override, Synthesis review, Telemedicine), and Institutional Administrators (Global telemetry, Audit logging).
- Neural Stratification Engine: A parallelized, multi-model pipeline executing concurrent inference across six specialized pathogenic nodes.
- Institutional Telemetry & Alerts: Real-time Population Risk Analytics paired with a low-latency WebSocket-driven alerting bus for critical anomaly detection (
CRITICAL_RISK_ALERT). - Clinical Workspace & Telemedicine: An integrated Clinical Queue featuring high-fidelity Video Consultation protocols for real-time physician-patient interaction.
- Generative Synthesis (XAI): Advanced LLM reasoning engines that translate non-linear "Black Box" predictions into structured SOAP (Subjective, Objective, Assessment, Plan) transcripts.
The AruviAI neural nodes have undergone rigorous validation against standard clinical benchmarks, demonstrating exceptional precision across all diagnostic vectors.
| Clinical Node | Algorithmic Methodology | Accuracy | Precision | Recall | F1-Score | AUC-ROC |
|---|---|---|---|---|---|---|
| Brain Tumor | EfficientNet-B0 (T-L) | 99.7% | 0.99 | 0.99 | 0.99 | 1.00 |
| Metabolic (Dia) | XGBoost Stratified | 89.1% | 0.88 | 0.91 | 0.89 | 0.93 |
| CVD (Heart) | Random Forest Ensemble | 87.3% | 0.85 | 0.89 | 0.87 | 0.91 |
| Cerebrovascular | Gradient Boosting | 85.7% | 0.83 | 0.87 | 0.85 | 0.89 |
| Renal (Kidney) | Logistic Regression + | 86.5% | 0.84 | 0.88 | 0.86 | 0.90 |
| Hepatic (Liver) | SVM Classifier | 84.2% | 0.81 | 0.86 | 0.83 | 0.87 |
Note
Bias & Generalization Constraints: The Brain Tumor Convolutional model has been validated for minimal generalization gap (-0.4%) and low-bias risk (0.01% class performance gap). Please consult the Bias & Fit Analysis for comprehensive validation metrics.
AruviAI mandates algorithmic transparency to maintain clinical trust:
- SHAP Vectoring: Local feature importance visualization for cardiovascular and metabolic predictions, granting clinicians direct insight into the localized drivers of a risk score.
- Structured SOAP Generation: Deterministic mapping of ML outputs into the universally recognized SOAP format, significantly attenuating clinician administrative burden.
- Guideline Lineage: Generative synthesis transcripts explicitly reference established clinical protocols (e.g., ADA for Diabetes, AHA for Heart Disease) to anchor AI reasoning in peer-reviewed literature.
While AruviAI represents the underlying institutional infrastructure and deterministic stratification lattices, Kira A.I. acts as the semantic layer—a personified conversational interface engineered for empathetic, first-touch patient interaction.
| Feature Matrix | AruviAI (Core Lattice) | Kira A.I. (Semantic Interface) |
|---|---|---|
| Logic Typology | Structured ML & Deep Learning Ensembles | Generative LLM (Conversational AI) |
| Primary Goal | High-precision diagnostic stratification | Empathic triage & Appointment orchestration |
| Output Vector | SOAP Transcripts, Risk Matrices, PDFs | Natural language dialogue, Intent extraction |
| User Role | Institutional oversight & Clinical Review | Direct user engagement & First-touch triage |
AruviAI introduces Provisio, a proprietary design system explicitly optimized for high-density clinical data presentation and institutional authority.
- Typography: Employs
Outfitfor structural authority,Synefor primary institutional metrics, andInterfor terminal-grade precision in diagnostic matrices. - Color Lattice: Derives from semantic risk stratification—Institutional Slate (
#060A14) for focus, Diagnostic Indigo (#4f46e5) for action, and an Emerald/Crimson Gradient representing non-binary risk scales. - Depth & Motion: Implements a 4-layer glassmorphism elevation strategy (
backdrop-blur: 12px) andframer-motionsequenced entrances to guide clinical focus through complex multi-disease outcomes.
Built on a robust, modern, and highly scalable enterprise stack.
| Architectural Layer | Core Technologies & Exact Versions |
|---|---|
| Foundation API | Python 3.11, FastAPI 0.110.0, SQLAlchemy 2.0.0, PostgreSQL 16 |
| Intelligence | Scikit-learn 1.6.1, PyTorch 2.0.0, OpenCV 4.8.0, SHAP 0.44.0 |
| Client Interface | React 19.2.0, Vite 7.2.4, TailwindCSS 4.2.1, Zustand 5.0.11 |
| UI Components | Radix UI, Framer Motion 12.26.2, React Three Fiber 9.5.0, Recharts 3.6.0 |
| Orchestration | Groq (Llama 3) 0.4.2, Google Gemini, WebSockets, JWT Authentication |
For academic researchers and system administrators looking to reproduce the stratification lattices locally:
# 1. Repository Synchronization
git clone https://github.com/Baskaran0402/HEALTHCARE_PROJECT.git
cd HEALTHCARE_PROJECT
# 2. Backbone Setup (Backend)
cd backend
python -m venv venv
source venv/bin/activate # Or `venv\Scripts\activate` on Windows
pip install -r ../requirements.txt
# 3. Intelligence OS Activation
python -m backend.main
# 4. Interface Hydration (Frontend)
# Open a new terminal window
cd frontend
npm install
npm run devHEALTHCARE_PROJECT/
├── backend/ # FastAPI Institutional Gateway
│ ├── routers/ # REST/WebSocket API Endpoints
│ └── main.py # Application Entrypoint
├── frontend/ # React 19 Intelligence Console
│ └── src/components/ui/ # Atomic Enterprise Components
├── src/ # Core ML/AI Neural Orchestration
│ ├── agents/ # Specialized Pathogenic Nodes
│ │ └── kira_agent.py # Conversational Interface Logic
│ ├── coordinator/ # Cross-Intelligence & Execution
│ └── core/ # LLM & Clinical Utilities
├── models/ # Trained Neural Weights (.pkl/.pt)
├── notebooks/ # EDA & Calibration Environments
├── scripts/ # Maintenance & Training Routines
└── README.md # Master DocumentationAruviAI is classified as a Clinical Decision Support Tool, not a definitive diagnostic replacement.
- Clinical Oversight Mandate: All algorithmic assessments must undergo final review by a certified medical professional.
- Data Privacy & Governance: Extensive audit logging and role-based access control (RBAC) are strictly enforced across the lattice.
- Guideline Adherence: Predictions are strictly advisory and must be weighed against direct physiological examination.
Baskaran S Lead Architect & AI Researcher
- LinkedIn: Baskaran S
- GitHub: @Baskaran0402
- Email: baskarseenu2005@gmail.com
If you incorporate this system or its multi-node stratification architecture into your academic research, please cite as:
@software{aruvi_ai_2026,
author = {Baskaran S},
title = {AruviAI: Integrated Clinical Intelligence & Multi-Node Diagnostic Stratification Lattice},
year = {2026},
url = {https://github.com/Baskaran0402/HEALTHCARE_PROJECT}
}
Built for institutional excellence. Strategic clinical intelligence.
🩺 AruviAI Professional
🩺 AruviAI Professional