VigilAI

Vigilant AI, Unyielding Protection

Comment

Inspiration

The core inspiration behind VigilAI comes from the concept of 'vigilance', the need for constant monitoring and protection in cybersecurity. With cyber threats becoming more sophisticated and harder to detect, there’s a clear need for AI-powered solutions that offer real-time, adaptive threat detection and automated response. By combining the power of AI and machine learning, VigilAI aims to emulate human-like vigilance while continuously learning and evolving to protect systems from emerging threats. The main benefit of using AI for this is Faster detection of cybersecurity attacks.

What it does

VigilAI is an AI based cybersecurity platform with multiple features.

  1. Malware File Detection using Siamese Networks: Detects if a file is Malware by comparing with a reference malware file.
  2. Cross Site Scripting (XSS) Vulnerability Detection : User can enter a peice of code and the model checks if the code is vulnerable to XSS or not.
  3. Email chatbot and phishing mail detector: User has to enter number of emails which will be accessed via Gmail API and RAG chatbot will be created on its basis. There is also a feature of AI detecting whether the email content is malicious or not.

How we built it

  1. Malware File Detection: This model is created by training DikeDataset, taken from GitHub. The dataset contains benign and Malware samples. A Siamese Network is trained on this dataset where we train the dataset in pairs. There is a reference file which is a Malware file, so more is the similarity score, more is the chance of being a malware. Architecture : The Siamese Network has 2 Convolution layers, 2 MaxPooling layers. The model is trained in 15 epochs and has 91% accuracy.

  2. Cross Site Scripting (XSS) Vulnerability Detection: This model is a LSTM model. The dataset is taken from Kaggle "Cross site scripting XSS dataset for Deep learning". Architecture : The model contains 2 LSTM layers, 1 Dense, 1 B.N and 2 Dropout layers with 97% accuracy.

  3. Email chatbot and phishing mail detector : Email chatbot is a RAG based chatbot where user enters a certain number of emails and Gmail API will access those and send them to DynamoDB. It will send details like text, sender etc along with embedddings created using OpenAI (text-embeddings-3-small) and the top embeddings and details will be retrieved based on similarity with query. Then, the top results will be sent to OpenAI for the response. The phishing mail will be detected by OpenAI.

Challenges we ran into

  1. Getting an enthical dataset for Malware Detection was difficult
  2. Training Siamese Network was difficult because of the amount of data.
  3. Integrating Gmail API to retrieve emails and process them in real time was tricky

Accomplishments that we're proud of

  1. Siamese Network for Malware Detection: Successfully training and deploying a Siamese Network for malware detection, achieving an impressive 91% accuracy, was a major milestone.

  2. XSS Detection Model: Developing an LSTM-based model for XSS vulnerability detection with 97% accuracy demonstrated the robustness of the deep learning approach for cybersecurity.

  3. Email Chatbot & Phishing Detection: Building a chatbot that uses email data from Gmail and creating an AI-based phishing mail detector are innovative steps toward automating email security.

What we learned

We learned how powerful deep learning techniques, like LSTM and Siamese Networks, can be for cybersecurity. We also learned the importance of high-quality labeled datasets for training accurate models

What's next for VigilAI

In the future, we would like to improve the project as much as possible with following features :

  1. Integrating Gmail API with the Malware Detection model
  2. Creating a Chrome extension apart from the Website, since that could be easily integrated with browser.

Built With

Share this project:

Updates