PhD Candidate in Informatics · Pennsylvania State University
Seeking Summer 2027 research and applied-science internships.
I am a PhD candidate in Informatics at Penn State, advised by Shomir Wilson in the Human Language Technologies Lab.
My research focuses on two complementary areas: agentic AI privacy—how LLMs, tools, and context shape privacy risks when agentic systems act on users’ behalf—and NLP for consumer-oriented legal documents—how language technologies can help consumers navigate contracts and policies, and the failure modes that can limit their usefulness.
AI systems increasingly mediate information between people, digital services, and complex documents. I study this role through two NLP-centered research areas: privacy in agentic AI systems and language technologies for consumer-oriented legal documents.
Agentic AI systems combine LLMs, tools, context, and system orchestration to act on users' behalf. I study how these components shape personal-information disclosure, data minimization, and privacy risks when agents communicate with third-party services.
Consumer-oriented legal documents—such as contracts, privacy policies, terms of service, and health care agreements—are often difficult to navigate and use. I study how NLP and LLM systems can help consumers find, compare, and understand consequential information in these documents, as well as the failure modes that can make those tools incomplete or misleading.
PrivaSeer is a large-scale search engine and research corpus designed to make website privacy policies transparent, searchable, and discoverable at web scale. [1] [2] [3]
I am a member of the PrivaSeer project team. My work contributes to sector classification and longitudinal measurement of disclosure patterns across online services.
Privacy, More or Less is under revision for PoPETs 2027.
Invited to give a guest lecture in IST 574: Human Language Technologies at Penn State.
Presented From Conventional Web Privacy to Agentic Disclosure at PrivateNLP @ ACL 2026.
SoACer published at ACM DocEng 2025.
PrivaSeer presented at USENIX SOUPS 2025.
PrivateNLP @ ACL 2026
ACM DocEng 2025 · Corpus of 195,495 websites
USENIX SOUPS 2025 · Poster · 3,967,487 policies as of April 2025
FLAIRS-35 · 2022
Ongoing research
Uses controlled schema interventions to test effects on unnecessary personal-information disclosure while holding the task and backend operation fixed.
Manuscript in preparation · Target: ACM FAccT 2027
Uses 24 topics and 3,899 verified provisions to evaluate allocation, omission, scope preservation, and factual support.
Under revision for PoPETs 2027
Analyzes 59,590 matched 2022–2025 policy pairs for cross-sector changes in recipient specificity, purpose coverage, and disclosure opacity.
Invited guest lecture (scheduled), IST 574: Human Language Technologies, Penn State.
Presented From Conventional Web Privacy to Agentic Disclosure at PrivateNLP @ ACL 2026, San Diego.
Participant, Mila Responsible AI and Human Rights Summer School, Montréal.