ScanCode-Toolkit : Mark required phrases for rules automatically using NLP/AI

ScanCode-Toolkit relies on "required phrases" to prevent false positive license detections, but 74% of its rules currently lack this protection. To solve this, I am building an AI/NLP pipeline that automatically reads raw license texts and extracts the exact phrases needed to protect these rules. I have already built a working proof-of-concept, complete with a trained model and a live web demo that the maintainers can test right now. During GSoC, I will build this into a lightweight, production ready tool. My core deliverables include a highly accurate machine learning model optimized for offline use and a native Command Line Interface (CLI). This tool will automatically inject high-confidence phrases and provide an interactive "human-in-the-loop" review mode for edge cases, ultimately delivering a measurable reduction in real world false positives.