Sandbox Code Execution System with Integrated AI-Driven Assistance

Author : Adithya V

Abstract : This paper presents a secure and scalable online code execution system that leverages Docker containers for robust sandboxing, utilizing Linux control groups (cgroups), namespaces, and seccomp profiles to ensure isolation and security. An AI-driven chatbot, powered by LLaMA 3.2, enhances user learning by providing context-aware programming guidance. User- submitted code is executed in isolated containers, with cgroups enforcing resource limits and seccomp profiles restricting system calls to mitigate threats like fork bombs, infinite loops, and container escape attempts. A Spring Boot backend with WebSocket communica- tion enables real-time interaction, while a PostgreSQL database logs execution metrics and user activity. The AI chatbot analyzes code and terminal output to offer tai- lored debugging and educational support. Performance evaluations show Docker containers achieve startup times of approximately 0.8 seconds, significantly faster than virtual machines (around 12 seconds), with additional advantages in memory, I/O, and energy efficiency. This study explores the system’s architecture, security mecha- nisms, AI integration, and performance, highlighting its suitability for educational and competitive programming environments.

Keywords : Docker, Code Execution System, code isolation, sandboxing.

Conference Name : International Conference on Information and Education Innovations (ICIEI - 25)

Conference Place : Gurgaon, India

Conference Date : 13th Dec 2025