# SafeGuard 🛡️

SafeGuard is an all-in-one, secure messaging platform designed to give kids a unified inbox while giving parents peace of mind. It aggregates messages from platforms like Discord, Instagram, Twitter, and WhatsApp into a single application. Behind the scenes, a powerful, custom-built AI filtering engine scans both text and images in real-time to blur and censor harmful content, cyberbullying, and NSFW media before the child ever sees it.

Parents have access to a dedicated dashboard featuring data visualizations, stranger-danger approval queues, and a "Privacy Mode" that respects the child's privacy by only showing flagged content.

---

🏗️ Architecture & App Sections

1. Frontend

The frontend is built for speed and responsiveness using React, TypeScript, and Vite. Styling is handled via Tailwind CSS.

• Child View (/child-chat): A unified chat interface resembling popular messaging apps where kids can talk to approved contacts across multiple platforms.
• Parent Dashboard (/parent-dashboard): A monitoring hub equipped with statistics (Safe Rate, 7-Day Activity), a queue for pending contact approvals, and real-time alerts for censored messages.
• Content Checker: A lightweight, in-browser regex checker runs on the child's input to warn them before sending sensitive information (like addresses or SSNs) or profanity.

2. Backend

The backend is a fast, asynchronous REST API powered by Python and FastAPI.

• Platform Integrations: It runs background tasks to poll and listen to platforms like Instagram (via instagrapi) and Discord (via discord.py-self).
• Message Processing: It ingests incoming messages, passes them through the AI filtering pipeline, uploads media attachments securely, and saves the final processed data to the database.

3. Models (AI / Machine Learning)

SafeGuard uses a robust, in-house, multi-layered machine learning pipeline for real-time moderation:

• Text Censoring: A custom-trained sequence classification model (using PyTorch and HuggingFace Transformers) detects nuanced inappropriate content, threats, harassment, profanity, slurs, and other harmful phrases.
• Image Moderation: An approach using a SigLIP2 classifier. This pipeline accurately detects any harmful images or NSFW content.

4. Database (Firebase)

The application relies on Firebase for its real-time database and storage needs:

• Cloud Firestore: Stores all messages, user settings, blocked users, and approved contacts. Messages are saved with detailed metadata, including whether they were censored and the specific words flagged.
• Firebase Storage: Securely hosts intercepted media attachments and profile pictures, allowing the frontend to load them safely.

---

📦 Requirements & Packages

Frontend

• Node.js (v16+) & npm
• Dependencies: react, react-dom, react-router-dom, firebase, @phosphor-icons/react, tailwindcss

Backend & Models

• Python (3.8+)
• Core Packages: fastapi, uvicorn, python-dotenv, firebase-admin.
• Platform Packages: discord.py-self, instagrapi.
• Machine Learning Packages: torch, transformers, numpy, pandas, scikit-learn, opencv-python, joblib, Pillow.
• Local LLM: Ollama installed locally with the llama3.2:3b model downloaded.

---

🚀 Setup & Installation Guide

Step 1: Clone the Repository

Clone the project to your local machine and navigate into the root directory.

Step 2: Firebase Configuration

1. Go to the Firebase Console and create a new project with Firestore and Storage enabled.
2. Generate a service account private key (JSON) and save it in the backend/ directory as serviceAccountKey.json.
3. Get your Firebase Web App configuration keys.

Step 3: Environment Variables

Create .env files in both the root/frontend and backend/ directories.

• Frontend (/.env): Add your Vite Firebase config (VITE_FIREBASE_API_KEY, VITE_FIREBASE_PROJECT_ID, etc.).
• Backend (backend/.env): Add your Discord Token, Instagram credentials, and Firebase paths.

Step 4: Frontend Setup

Open a terminal in the root directory:

# Install frontend dependencies
npm install

# Start the Vite development server
npm run dev