Aigis is an AI-powered browser extension and dashboard designed to protect children and families from online dangers. It leverages advanced machine learning and deep learning models for real-time detection of harmful content, deepfakes, and risky interactions. This project was developed as part of the AI Challenge based Learning (CBL) Project at ESPRIT University.
Web Extension Protecting children and families from online dangers using AI-powered detection and supervision tools
Children are increasingly exposed to online threats:
- 👤 66% of kids chat daily with strangers online — risking deepfakes and grooming.
- 💬 12% of parents report adults trying to befriend their children online.
- 🎮 54% of kids encounter violent content — some at risk of photosensitive epilepsy.
- 📹 Deepfakes make fake audio/video harder to detect during cam interactions.
- 🎮 Most online games are not approved by safety regulations, unlike AAA games.
- 📱 Kids aged 8–12 spend 5.4 hrs/day, teens 8.2 hrs/day online — often unsupervised.
Aigis is our AI-powered browser extension and dashboard to detect threats and protect families in real time, utilizing Python, Django, JavaScript, Machine Learning (ML), and Deep Learning (DL).
- 🔍 AI-Powered Chat Supervision: Utilizes Natural Language Processing (NLP) models like RoBERTa (Transformers) and the Gemini API to detect predator-like behavior, offensive language, hate speech, and other harmful patterns in text messages. Keywords:
text-classification,nlp,chat-analysis,content-moderation. - 🧑🚀 Fake Profile & Deepfake Detection: Employs deep learning (TensorFlow/Keras) and computer vision (OpenCV, Ultralytics) techniques for deepfake analysis in images/videos and behavior pattern scanning to identify suspicious profiles. Keywords:
deepfake-detection,image-analysis,video-analysis,biometric-security. - ⚔️ Violent Content & Game Flagging: Analyzes visuals (OpenCV) and content tags for violence, and detects flashing lights (NumPy, OpenCV) to mitigate epileptic risks in games and videos. Keywords:
violence-detection,epilepsy-protection,image-recognition,video-processing. - 🗣️ Voice/Video Analysis: Leverages audio processing (Librosa, SoundFile, Pydub) and computer vision (OpenCV) to detect fake audio, dangerous or inappropriate language, and harmful visual content in live cam discussions. Keywords:
voice-analysis,speech-recognition,audio-processing,real-time-analysis. - 🛡️ Parental Dashboard: Provides a centralized web interface (built with Django and potentially enhanced with tools like Streamlit for data visualization) for parents to review alerts, detailed reports, and safety statistics generated by the Aigis extension. Keywords:
parental-controls,dashboard,reporting,data-visualization. - 🤖 Explainable AI (XAI): Potentially incorporating tools like LIME for understanding model predictions, enhancing trust and transparency. Keywords:
explainable-ai,model-interpretability.
- Languages:
HTML5,CSS3,JavaScript (ES6+) - Framework/Libraries: Vanilla JS (for core logic, content scripts, popup scripts)
- Tools:
html2canvas(for webpage screenshot functionality) - APIs: Chrome Extension APIs (runtime messaging, storage, alarms)
- Styling: Custom CSS (No Tailwind CSS or other CSS frameworks)
- Core Framework:
PythonwithDjango&Django REST Framework(for robust API endpoints) - Asynchronous Operations:
ChannelsandDaphne(for WebSocket/real-time communication if needed) - Networking:
requests(for external API calls, e.g., Gemini API),django-cors-headers(for Cross-Origin Resource Sharing)
AI Models & Detection Libraries:
- Text Analysis (NLP):
Transformers (Hugging Face): For RoBERTa-based text classification (Offensive, Hate, Safe).Google Gemini API: For advanced thematic text analysis (Manipulative, Suicide Risk, Blackmail, Meeting Attempts).Sentence-Transformers: For generating text embeddings.
- Image/Video Analysis (Computer Vision):
OpenCV: Core library for image/video processing, violence detection, flashing lights detection.Pillow (PIL): Image decoding, manipulation, and preprocessing.TensorFlow / Keras: Custom-trained deep learning models for "Brainrot" content detection and Deepfake detection.Ultralytics (YOLO): Potentially for object detection or other advanced vision tasks.
- Audio Analysis:
Librosa: For advanced audio signal analysis and feature extraction.SoundFile,Pydub: For audio file manipulation and processing.
- General Machine Learning & Numerics:
NumPy: Fundamental package for numerical computation, used extensively in CV and ML tasks.ONNXRuntime: For running models in the Open Neural Network Exchange format, enabling cross-platform compatibility.scikit-learn(implicitly via other libs or for general ML tasks if used)
- Model Explainability:
LIME: For Local Interpretable Model-agnostic Explanations.
- Data Handling & Orchestration:
LangChain,LangChain-Community,LangChain-Groq,LangChain-Huggingface: For building applications with LLMs, potentially RAG (Retrieval Augmented Generation) for knowledge base interactions.Faiss-CPU: For efficient similarity search, often used with embeddings in RAG systems.Pydantic(often a dependency of LangChain/FastAPI-like projects, good for data validation)
- ORM:
Django ORM - Default:
SQLite(for ease of setup and development) - Production-Ready: Easily swappable with
PostgreSQLorMySQL
- Environment Management:
python-dotenv(for managing environment variables) - Version Control:
Git,GitHub - Virtual Environments:
venv - Notebooks/Experimentation:
JupyterLab,ipywidgets(for interactive development) - Data Visualization/Demoing:
Streamlit(available for building interactive data apps)
- 👤 Samar Souissi – LinkedIn
- 👤 Med Aziz Maatoug – LinkedIn
- 👤 Linda Sakouhi – LinkedIn
- 👤 Fedi Abassi – LinkedIn
- 👤 Med Anas Obba – LinkedIn
- 👤 Taher Ayadi – LinkedIn
Built with ❤️ as part of our university AI CBL Project to bring safety, ethics, and AI together.
git clone https://github.com/SynapsysESPRIT/AIGIS.git
cd AIGISpython -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txtImportant: After activating your environment, set your API keys and tokens as environment variables:
# For Linux/macOS (bash/zsh)
export HUGGINGFACE_TOKEN="your_huggingface_token_here"
export GROQAPI_KEY="your_groq_api_key_here"
export ELEVENLABS_API_KEY="your_elevenlabs_api_key_here"
export GEMINI_API_KEY="your_gemini_api_key_here"
# For Windows (Command Prompt)
set HUGGINGFACE_TOKEN=your_huggingface_token_here
set GROQAPI_KEY=your_groq_api_key_here
set ELEVENLABS_API_KEY=your_elevenlabs_api_key_here
set GEMINI_API_KEY=your_gemini_api_key_here
# For Windows (PowerShell)
$env:HUGGINGFACE_TOKEN="your_huggingface_token_here"
$env:GROQAPI_KEY="your_groq_api_key_here"
$env:ELEVENLABS_API_KEY="your_elevenlabs_api_key_here"
$env:GEMINI_API_KEY="your_gemini_api_key_here"cd Backend
python manage.py migrate
python manage.py runserver- Open Chrome and go to
chrome://extensions/ - Enable "Developer mode"
- Click "Load unpacked" and select the
extensionfolder
- The backend will be available at
http://127.0.0.1:8000/ - The extension popup will be available in your browser toolbar after loading.
- Make sure the back 79BE end is running before using the extension for detection features.
SynapsysAI Children’s Online Protection License (SCOPL) - v1.0 Copyright (c) 2025 SynapsysAI™
Permission is hereby granted to use, copy, and modify this software for non-commercial and educational purposes only, subject to the following conditions:
This software is intended for the protection of children online and may not be used for surveillance, exploitation, or any unethical purposes.
All copies or substantial portions of the software must include this notice.
Commercial use, resale, or redistribution without prior written permission from Synapsys is strictly prohibited.
The name "SynapsysAI" and associated branding may not be used without explicit written permission.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND. UNDER NO CIRCUMSTANCES SHALL SYNAPSYS OR ITS CONTRIBUTORS BE LIABLE FOR ANY DAMAGE RESULTING FROM THE USE OR MISUSE OF THIS SOFTWARE.