Detecting software vulnerabilities using large language models

Software vulnerability is a huge problem in the modern world, but detecting such vulnerabilities is becoming increasingly time-consuming and resource dependent, especially in large-scale projects. To address this, our team has developed a software vulnerability detection tool with a fine-tuned Large Language Model (LLM) integrated to quickly track and analyze vulnerabilities within program codes.

Users can input raw code, code file or Github repository link for the application, and it will quickly return results with high accuracy. Users can also sign-in to track their past code submissions and join in the Community Board to help other users or seek help for themselves.

The system can also generate new vulnerability data based on users’ posts for further fine-tuning on a fixed interval, which can further increase our scalability and adaptability since vulnerability in codes are endless and new ones will always be found after a short period of time. With this application, we hope to pave the way for future AI-driven tools and ease the process of software vulnerability detection.