AI Website Navigation Assistant

This is an AI-powered Chrome Extension that helps users navigate complex websites. It acts as a smart sidecar, allowing users to ask "Where is X?" or "How do I do Y?" and getting direct navigation assistance.

It features a dual-mode AI backend:

1. Cloud Mode: Uses Google Gemini (via Node.js) for high-intelligence reasoning and general queries.
2. Local Mode: Uses a locally hosted SentenceTransformer model (via Python) for offline, privacy-focused, and fast semantic link matching.

✨ Key Features

• Dual AI Modes: Toggle between Cloud (Gemini) and Local (Python) AI.
• Voice Commands: Click the microphone to speak your query.
• Smart Shortcuts:
  • Shift + F: Toggle the chat window (won't trigger while typing).
  • Escape: Close/Hide the chat window.
  • bot exit: Type this command to close the chat.
• Link Highlighting: The assistant highlights the relevant link on the page before clicking it.
• Local Model Fine-Tuning: Train the local AI on your own custom data for better accuracy on specific sites.
• Firewall (Privacy Shield): Automatically blocks third-party trackers (ads, analytics) like doubleclick and google-analytics while keeping banking/login sites safe.
• Optimized Local Navigation: Intelligent heuristics to instantly understand commands like "change password" or "edit settings" without wait times.

🛠️ Installation & Setup

1. Clone the Repository

git clone <repository-url>
cd Hackathon-Hacktide

2. Chrome Extension

1. Open Chrome and navigate to chrome://extensions.
2. Enable Developer Mode (top right).
3. Click Load unpacked and select the Hackathon-Hacktide folder.

3. Backend Servers

You need to run both servers for full functionality.

Node.js Server (Cloud Mode)

cd server
npm install
# Create a .env file with GEMINI_API_KEY=your_key
node server.js

Runs on port 3000.

Python Server (Local Mode)

cd server
pip install flask flask-cors sentence-transformers
python ai_server.py

Runs on port 5000.

🧠 Local Model Training

You can fine-tune the local AI model to understand specific jargon or website mappings.

1. Edit Data: Open server/training_data.json and add your examples:

   {
     "query": "Where can I see my grades?",
     "positive": "Academic Reports",
     "negative": "Hostel Fee"
   }

2. Train: Run the training script.

   cd server
   python train.py

3. Restart: Restart ai_server.py to load the new model.

📂 Project Structure

• manifest.json: Chrome extension configuration.
• content.js: Main logic injected into webpages (UI, event listeners).
• styles.css: Styling for the chat interface.
• server/: Backend code.
  • server.js: Node.js server for Gemini API.
  • ai_server.py: Python server for local embeddings.
  • train.py: Script to fine-tune the local model.

🚀 Usage

1. Open any webpage.
2. Press Shift + F or click the 🤖 icon to open the assistant.
3. Type your question (e.g., "Login page", "Contact support").
4. The AI will find the best link, highlight it, and navigate for you.