1st Position at Gajshield KJSCE Hack 8

Date: 12th & 13th April 2025
Achievement: 1st Position (24-hour Hackathon)

Project: Kavach.AI

Kavach.AI was built during a hackathon to explore a better way to detect and analyze malware. Our goal was to create a working system that could safely run suspicious files, study how they behave, and give useful insights all in one place. We set up a secure Docker-based environment to test real malware and collect system call data for deeper analysis.

Technical Approach

To detect threats, we used an XGBoost machine learning model trained on malware samples, which gave us good accuracy. We also added a simple method to check file entropy, which helps spot hidden or obfuscated files. For system behavior, we flagged unusual calls and used a language model (LLama 3.2) to help interpret what might be going wrong.

In the end, we built a basic but functional prototype that combines static and dynamic analysis on a Desktop Application using Electron JS. It showed potential to grow further, and we noted future steps like real-time detection and integration with existing security tools.

Our application was tested by the judges of the hackathon from Gajshield with their own malware files and our application was a success.

Tech Stack

• Frontend/Desktop: Electron.JS
• Virtualization: Docker
• Backend/AI: Python, Pytorch, Flask
• Version Control: Git, Github