Research focus
M3P is a security-oriented pre-processing pipeline designed to reduce sensitive-data exposure and unsafe prompt behaviour before content is passed to a language model.
Pipeline
The architecture combines four stages: Masking, Prompt Adjustment, Policies, and Pre-Check. The design emphasizes auditability and a clear verdict path rather than treating safety as a single opaque classifier.
Model work
The masking component explores token classification for personally identifiable information and software secrets in text and code-oriented contexts. The wider workflow combines learned and deterministic components rather than forcing every problem into one model.
Why it belongs in the portfolio
It demonstrates applied security research, machine-learning integration, prompt security thinking, data protection and system design in one end-to-end problem.