I’ve been building a small ML project that tries to detect vulnerabilities in source code. It uses a real dataset (Big-Vul) and a simple TF-IDF + Random Forest model to classify code as safe or vulnerable. After fixing a lot of dataset and file path issues, I finally got the full pipeline working end-to-end with around 94% accuracy on the test split. Most of the work so far wasn’t even the model itself, but getting the data loading and training pipeline stable. Next I want to turn it into a real tool that can scan files directly, highlight risky lines, and eventually improve the model to detect different types of vulnerabilities.
