🏥 Vertice AI Health (Humanitarian Edition)

A "High-Tech, Low-Resource" Clinical Decision Support System (CDSS) for humanitarian zones. Designed for 2026. Built with Google Gemini 3 Pro.

Features • Architecture • Getting Started • Philosophy

---

🌍 The Mission

In humanitarian crisis zones, doctors face language barriers, extreme exhaustion, and low patient health literacy. Vertice AI Health is not just a chatbot; it's an Agentic Operating System designed to operate in low-bandwidth environments, acting as a super-specialist team in the pocket of a field doctor.

🚀 Key Features

1. 🧠 Trinity Diagnosis Protocol™

We don't rely on a single LLM call. We instantiate three distinct cognitive personas that analyze the patient in parallel, followed by a "Judge" agent that fuses the insights using Adversarial Review.

• 🛡️ Dr. Conservative: Focuses on safety, standard protocols (WHO/MSF), and "do no harm".
• 🔬 Dr. House (Aggressive): Hunts for rare diseases, "zebras", and hidden correlations.
• 🎓 Prof. Academic: Analyzes pathophysiology and evidence-based likelihood ratios.
• ⚖️ The Judge: A temperature-0 agent that calculates a Confidence Score. If < 60%, the diagnosis is blocked for safety.

2. 🎙️ Vox Medicus (Multimodal Scribe)

No typing required. Using Gemini 3 Pro's native multimodal capabilities and the Opus codec for ultra-low bandwidth:

• Listens to the consultation audio directly.
• Analyzes acoustics (detects dyspnea, pain in voice, anxiety).
• Auto-fills complex clinical forms (PatientData JSON) from unstructured speech.

3. 🎨 Generative Pictogram Prescriptions

Solving the Health Illiteracy crisis.

• Instead of text instructions ("Take 2 pills every 8h"), the system generates a Visual Recipe.
• Uses a React SVG Engine driven by Gemini's structured output to render offline-capable pictograms (Sun, Moon, Pill quantities) that any patient can understand regardless of language.

4. 🛰️ Telemedicine Hub (Google Stack)

• Smart Referral: Generates <100 word summaries optimized for WhatsApp/Satellite text.
• Native Meet Integration: One-click seamless video handoff to specialists.

---

🏗️ Trinity Architecture

The system runs Client-Side (Serverless) to ensure privacy and zero infrastructure costs for NGOs.

graph TD
    A[Doctor Input (Voice/Text/Img)] --> B{Vox Medicus}
    B --> C[Structured Patient Data]
    C --> D[Trinity Router]
    
    D --> E[Persona: Conservative]
    D --> F[Persona: Aggressive]
    D --> G[Persona: Academic]
    
    E & F & G --> H[The Judge (Consensus Engine)]
    
    H -- Score < 60% --> I[⛔ Safety Block Protocol]
    H -- Score > 60% --> J[✅ Clinical Report]
    
    J --> K[Generative Pictograms (Patient View)]

Loading

🛠️ Tech Stack (2026 Standard)

• AI Core: Google Gemini 3 Pro (via @google/genai SDK).
• Frontend: React 19 + Vite (TypeScript).
• Styling: Tailwind CSS (Medical Liquid Glass Theme).
• Audio: MediaRecorder API with Opus Codec (WebM).
• Icons: Lucide React (Dynamic Rendering).

⚡ Getting Started

Prerequisites

• Node.js 20+
• A Google Gemini API Key (Vertex AI or AI Studio)

Installation

1. Clone the repository

   git clone https://github.com/JuanCS-Dev/vertice-health.git
   cd vertice-health

2. Install dependencies

   npm install

3. Configure Environment Create a .env file in the root:

   VITE_GEMINI_API_KEY=your_gemini_api_key_here

4. Run Development Server

   npm run dev

🛡️ "Padrão Pagani" & Constitution

This codebase adheres to the Constituição Vértice v3.0:

• Zero Placeholders: No // TODO in critical paths.
• Type Safety: Strict TypeScript interfaces for all AI inputs/outputs.
• Sovereignty of Intent: The AI never "fakes" a diagnosis. If it doesn't know, it says so.

---

🤝 Contributing

This is an open-source project for the Google Gemini Hackathon. Pull requests focused on accessibility, offline-first capabilities, and localization are welcome.

License: MIT Author: JuanCS-Dev (Executor Tático)