OncoVector AI

Clinical Decision Support with Grounded Reasoning & Vector Retrieval

OncoVector is a next-generation healthcare application designed to reduce diagnostic errors by grounding AI reasoning in real-world university case registries. It utilizes a "Sequential Neural Pipeline" to analyze patient data, compare it against high-fidelity medical datasets (like NIH DeepLesion), and provide evidence-based risk assessments.

🚀 Key Features

• Multimodal Analysis: Ingests clinical notes and patient imagery (CT, MRI, Dermoscopy).
• Vector Space Retrieval: Simulates semantic search against federated datasets (NIH, TCIA, ISIC).
• Explainable AI: Provides a "Reasoning" view explaining why a specific diagnosis was suggested.
• Demo Mode: Fully functional UI simulation without requiring live API credentials.

🛠️ Tech Stack

• Frontend: React 19, TypeScript, Tailwind CSS
• Visualization: Recharts, D3.js
• AI/Backend: Google GenAI SDK (gemini-3-pro-preview, gemini-2.5-flash)
• Icons: Lucide React

---

🏁 Getting Started

You can run this application in two modes: Live Mode (connected to Gemini API) or Demo Mode (simulated data).

Prerequisites

• Node.js (v18 or higher)
• npm or yarn

Installation

1. Clone the repository:

   git clone https://github.com/your-username/oncovector-ai.git
   cd oncovector-ai

2. Install dependencies:

   npm install

🟢 Running in Demo Mode (No API Key Required)

The application is architected to detect missing credentials and automatically switch to Demo Mode. This allows you to explore the UI, animations, and data visualization flow without a Google Cloud account.

1. Simply start the development server:

   npm start

2. Open http://localhost:3000 in your browser.
3. Fill in the patient intake form (or leave defaults) and click "Run Multi-Modal Analysis".
4. The system will simulate the pipeline steps and generate a pre-calculated, high-fidelity report.

🔴 Running in Live Mode (With Gemini API)

To enable real AI reasoning and live web grounding:

1. Obtain an API Key from Google AI Studio.
2. Create a .env file in the root directory (or set the variable in your environment):

   API_KEY=your_actual_api_key_here

3. Start the application:

   npm start

---

📂 Project Structure

/
├── components/          # UI Components (Graphs, Modals, Network Viz)
├── services/
│   ├── geminiService.ts # AI Logic (Handles Live & Mock Fallbacks)
│   └── mockVectorDb.ts  # Simulated Vector Database Registry
├── types.ts             # TypeScript Interfaces
├── App.tsx              # Main Application Controller
└── index.tsx            # Entry Point

🧠 How It Works (Architecture)

1. Input: User inputs patient age, symptoms, and uploads a scan.
2. Vision Layer: The app (simulates or performs) feature extraction to identify the anatomy (e.g., Skin vs. Lung).
3. Vector Retrieval: It queries a mock index of 12 real-world datasets (like TCIA-GBM or ISIC Melanoma) to find "Nearest Neighbors".
4. Reasoning Synthesis:
   • Live Mode: Gemini 3 Pro synthesizes the patient data + retrieved cases to form a diagnosis.
   • Demo Mode: Returns a structured "Melanoma" or "Carcinoma" example case to demonstrate UI capabilities.
5. Output: Displays risk scores, comparison graphs (Radar/Scatter), and a clinical action plan.

---

📄 License

This project is open-source and available under the MIT License.