🚑 MediPatient AI

Hybrid Clinical Intelligence: ML + LLM Reasoning (2026 Version)

A research-grade clinical decision support system combining Random Forest ML for statistical detection with Gemini LLM for clinical reasoning, explainability, and treatment planning.

---

🧠 System Overview

Double-Check Architecture

Analytical Engine (ML)

• Random Forest trained on real hospital data + synthetic textbook cases.
• Feature engineering includes vitals + derived flags (fever_high, tachycardia, hypotension, hypoxia).
• Outputs preliminary diagnosis with confidence scores.

Reasoning Engine (LLM)

• Google Gemini 1.5 Flash validates ML predictions.
• Provides human-readable clinical reasoning, red flags, and differential diagnosis.
• Generates structured treatment plans.

Hybrid Orchestrator

• MedicalOrchestrator dynamically combines ML + LLM outputs.
• Intelligent override ensures Gemini corrects vague or incorrect ML predictions.
• Severity and urgency scoring based on evidence-based rules.

---

🏗️ Architecture Pipeline

1. Patient Data Intake

   • Flask API receives structured patient data from frontend (PatientFeatures).

2. ML Prediction

   • Random Forest outputs disease category and confidence.

3. LLM Validation

   • Gemini AI reviews ML output and applies overrides if needed.

4. Treatment Planning & Reasoning

   • Clinical reasoning, differentials, red flags, and treatment plan included.

5. Data Storage & Reporting

   • JSON patient records saved.
   • PDF reports generated and downloadable.

---

📂 Project Structure

File / Folder	Role
api.py	Flask server exposing secure endpoints for patient data, predictions, and report downloads.
AI_diagnosis.py	Hybrid ML trainer (real + synthetic data).
gen_ai_module.py	Gemini LLM integration (validation & treatment planning).
medical_orchestrator.py	Hybrid intelligence orchestrator combining ML + LLM outputs.
report_generator.py	Generates PDF clinical reports.
test_system.py	Validation & stress-testing of the pipeline.
models/	Trained ML model artifacts (.pkl, feature names, disease mapping).
.env	Gemini API key, model/report directories, optional rate limits.
templates/	Frontend dashboard (HTML).
static/	CSS/JS frontend assets.

---

🚀 Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Add .env configuration

GEMINI_API_KEY=your_key_here
MODEL_DIR=./output
REPORTS_DIR=./output/reports
PATIENT_DIR=./output/patient_data
ALLOWED_ORIGINS=http://localhost:3000,http://localhost:5000
FLASK_ENV=development

3. Train ML model (optional)

python AI_diagnosis.py

4. Start API

python api.py

5. Example request to /api/predict

POST /api/predict
Content-Type: application/json

{
  "age": 58,
  "gender": "male",
  "temperature": 37.2,
  "heartRate": 95,
  "systolicBP": 150,
  "diastolicBP": 95,
  "respiratoryRate": 20,
  "oxygenSaturation": 96,
  "painScore": 9,
  "symptoms": ["chest pain", "fatigue", "nausea", "dizziness"],
  "medical_history": ["hypertension"],
  "allergies": [],
  "medications": ["aspirin"]
}

---

⚙️ Features

• Hybrid AI: ML + LLM reasoning.
• Augmented Training: Synthetic cases complement real hospital data.
• Explainable AI: Full clinical reasoning, differentials, red flags.
• Intelligent Overrides: Gemini corrects vague/incorrect ML predictions.
• Treatment Planning: Structured plans generated dynamically.
• Severity & Urgency: Evidence-based scoring and triage.
• Secure API: Rate-limited, CORS-restricted, and path-traversal safe.
• PDF Reports: Downloadable via /download/.
• Patient Data Storage: JSON records with retrieval and deletion endpoints.

---

📡 API Endpoints

Endpoint	Method	Description
/api/predict	POST	Hybrid ML+LLM diagnosis, returns PDF link and reasoning.
/download/<filename>	GET	Download report PDF securely.
/api/status	GET	Server health check & model/LLM availability.
/api/get-reports	GET	List latest PDF reports.
/api/delete-report/<filename>	DELETE	Delete report PDF.
/api/get-patients	GET	List latest patient JSON records.
/download-patient/<filename>	GET	Download patient JSON securely.
/api/delete-patient/<filename>	DELETE	Delete patient JSON record.

---

🧪 Testing

python test_system.py

---

⚖️ Disclaimer

• Research Prototype: Not intended for real-world clinical diagnosis.
• **Use only under professional supervision.

---