San Diego Supercomputer Center, UC San Diego (REHS)
Research Intern — PII Detection in Support Tickets
Jun 2026 – Jul 2026
San Diego, CA · Mentor: Dr. Martin Kandes
SDSC’s staff get a steady flood of support tickets, many from researchers who aren’t systems people. A model that learned from years of past answers could help with that. The catch is that those tickets are full of names, emails, usernames and other personal details, so before any model sees them the personal information has to be found and removed.
That was my project in SDSC’s Research Experience for High School Students program, on a 7-person team mentored by Dr. Martin Kandes. I built an AI-assisted labeling tool to annotate the tickets, then benchmarked three ways of finding the PII across 1,400+ of them:
- Rule-based: Microsoft Presidio’s recognizers.
- A custom spaCy NER model trained for this kind of ticket text.
- LLM-based detection.
REHS overlapped with the start of my RISE internship in Boston, and it led straight into my current SDSC work on agents for researchers.