Ph.D. thesis titled "Controllable Personalization for Information Access" (defended May 2024).
Developed key datasets: LAMP, CSFCube, MSPT for under-explored problems.
Proposed methods/models including LACE, CtrlCE, Pearl, EdTM, Aspire, MINT—covering explainability, uncertainty calibration, inference-time feedback, and data-centric training.
Research includes interactive personalized ranking for peer review, literature brainstorming tools for scientists, retrieval-augmented personalized LLMs for writing, and low-resource IE for materials synthesis automation.
Conducted human-centered studies on lit-review behaviors and user-LLM collaboration.
Released pre-print in May 2025 analyzing user collaboration with LLMs like Bing Copilot and WildChat.
Presented at COLM 2024 (Philadelphia) and EMNLP 2024 (Miami).