Integrating Prompt Structures Using LLM Embeddings for Cybersecurity Threats
Kavya Nikhita Meda, Pavan Subhash Chandrabose Nara, Svoboda Bozenka, Tarek Zormati, Seth Turner, Wayne Worley, Reshmi Mitra · 2025
This paper aims to develop a specialized Large Language Model (LLM) for cybersecurity training, designed to educate users on fundamental cybersecurity concepts. This paper focuses on creating an interactive system where users can ask questions about computer security and receive accurate, informative responses. By addressing cybersecurity as a critical national issue, the LLM empowers individuals and organizations to defend against malicious cyber threats. Our system was developed using Python, utilizing Google Sheets as a database, Gradio for the user interface, and Google Gemini's API for advanced language processing. The implementation followed a test-driven development approach, iterating between coding and testing to ensure functionality and reliability. Key technologies include Mistral's Large 2 model and embedding models for clustering related data. The Retrieval-Augmented Generation (RAG) framework was employed to integrate information retrieval with the LLM, enhancing its accuracy and relevance. Tools such as Google Suite, Colab, and Gradio contributed to creating a robust and user-friendly system. This paper highlights the potential of domain-specific LLMs, offering a practical solution to the growing need for accessible cybersecurity education and fostering awareness to mitigate the risks posed by malicious hackers.