Kevin(Jian) Lu
Master's Student · Duke University
I am a Master's student in Computer Science at Duke University, where I am fortunate to be advised by Professor Anru Zhang. My research centers on efficient AI systems, across the stack:
- Efficient Inference — KV-cache optimization and sparse attention: understanding which parts of a model's computation can be compressed or skipped, and why some models are far more sensitive to it than others.
- Efficient Agentic Systems — agent design, memory, and low-latency reasoning, such as text-to-SQL agents that reason over databases with fewer, simpler steps.
These directions grew out of my work applying LLMs to AI for health, where efficiency is what makes powerful models practical to deploy in real clinical settings.
Previously, I obtained my B.S. in Computer Science from the University of Rochester, where I worked with Professor Jiebo Luo on computer vision methods for biomedical image analysis, and with Professor Christopher Kanan on evaluating multimodal large language models.
Selected Publications
Preprint
arXiv, 2026
Iteration Without Elaboration: A Simple ReAct Architecture Suffices for Text-to-SQL Generation
Preprint · Under Review
Submitted to JAMIA
Toward Automated Detection of Documentation Inconsistencies in Electronic Health Records
2026
IEEE International Conference on Healthcare Informatics (ICHI)
Reliable Curation of EHR Dataset via Large Language Models under Environmental Constraints
Education
2025 — Present
M.S. in Computer Science
Duke University, Durham, NC
Advisor: Prof. Anru Zhang. Research on efficient AI systems (efficient inference and agentic systems) and AI for health.
2021 — 2025
B.S. in Computer Science
University of Rochester, Rochester, NY
Worked with Prof. Jiebo Luo (2023 — 2025) on computer vision methods for biomedical image analysis, and with Prof. Christopher Kanan (2024 — 2025) on evaluating multimodal large language models.