GlossAPI: ML-assisted Anonymization Layer and Targeted Pipeline Improvements for Greek Datasets

Publicly available Greek datasets like those on OpenGov are a goldmine of information but they often contain personal details that need to be hidden to comply with privacy laws like the GDPR. Doing this automatically for Greek text is tough because the language is highly inflected and the documents are usually full of OCR scanning errors. To tackle this I am proposing a standalone ML assisted anonymization tool that plugs right into the glossAPI pipeline. Instead of a heavy slow model I am using a lightweight hybrid approach. It uses fast regular expressions and custom dictionaries to catch standard info and fix OCR typos while leaning on Microsoft Presidio and the spaCy multilingual pipeline to smartly detect complex entities like names and organizations. By wrapping this engine in a simple FastAPI REST interface we keep the main glossAPI codebase clean and independent. By the end of the project I will deliver this fully functional privacy API the necessary code to link it to glossAPI and clear practical documentation so future maintainers can easily protect sensitive public data.