Project insyt: Edge ML for Network Security

Our BYU Capstone team won 1st Place at the 2024 BYU Capstone Celebration in Machine Learning for our work with Sandia National Laboratories. We developed insyt, an AI-powered security system that treats log data as natural language.

The Technical Solution

We built a cutting-edge classifier capable of identifying cybersecurity attacks within massive log streams with high accuracy using a student-teacher distillation framework.

Tech Stack:

• Model Architecture: Developed using PyTorch and DistilBERT.
• Orchestration: Built a production-ready pipeline using Redis for queuing and SQLite for state management.
• Frontend: A React UI for real-time monitoring and visualization.
• Deployment: Released as a public Python package (pip install insyt) with an open-source model available on Hugging Face.

Impact

This project demonstrated that advanced NLP models can be made lightweight enough for edge security monitoring without sacrificing accuracy. It remains a key example of how model distillation can solve real-world cybersecurity challenges.