Exploring anonymization approach for GlossAPI GSoC idea

[raja-jaloka]

Dear Mentors ((@nikostsekos @myrsiniioannou @jimmmyss),

Hi, I’m Yuvraj, a student at KIIT University, India, pursuing a B.Tech in CSE and exploring the ML-assisted Anonymization Layer idea for GSoC’26. I’ve experimented with a small prototype(https://github.com/raja-jaloka/Hugging-Face-NER-Anonymizers-Regex-etc-GSOC-26-Practice-) combining:
1.Regex-based masking for structured identifiers (email, phone)
2.NER-based masking for semantic entities (names, orgs, locations)

I am also optimizing my prototype with GlINER, also exploring and working on my prototype with models like XLM-R,mT5,ByT5 to choose which one i should integrate finally for the project, it would be wonderful if you can point to where i should proceed?

The idea is to separate deterministic masking from ML-driven recognition before integrating into a dataset processing pipeline.
Currently I’m trying to understand how anonymization is envisioned within GlossAPI plus would it be more suitable as a standalone preprocessing stage or integrated directly into the existing pipeline?

Much appreciated thanks..