I study efficient database management and large language model (LLM) agents. My work includes LLM agents for program verification, learning-based query optimization, principled influence attribution in networks, and representation-aware data repair. I currently work on agent planning and memory management.
Outside of my Ph.D. research, I spent the summers of 2025 and 2026 at Uber, developing deep learning methods for marketplace forecasting and structural pricing.
Previously, I studied in the Economics and Computation master's program at Duke University before transitioning to the Ph.D. program. I received a bachelor's degree in Finance from Central University of Finance and Economics.
Whiel: A Neurosymbolic Platform for Certified Relational Program Verification
Jesse Comer*, Fangzhu Shen*, Shuo Zhang*, Mayur Naik, Sudeepa Roy, Val Tannen
Preprint, 2026. [pdf]
Combines LLM-generated invariants with Lean 4 and Vampire to verify relational programs, including recursive query equivalence, containment, and view-based rewriting.
Selectivity Estimation for Linear Queries via Online Learning Fangzhu Shen, Debmalya Panigrahi, Sudeepa Roy
Preprint, 2026. [pdf]
We recast selectivity estimation as an online learning problem, where both the data and the query workload may change adversarially and no distributional assumptions are needed. We give nearly tight regret bounds for point, range, subset, and linear queries over static and dynamic databases, and show a polynomial separation between squared and absolute loss.
Efficient Shapley-Based Influence Attribution in Social Networks Fangzhu Shen, Amir Gilad, Sudeepa Roy
ACM Conference on Knowledge Discovery and Data Mining (KDD), 2026. [pdf] [code]
We introduce a framework for ex-ante influence attribution: dividing credit among seed users before diffusion occurs using Shapley values from cooperative game theory. We give polynomial-time algorithms for single-step activation, prove #P-hardness for any longer propagation, and develop approximation algorithms with provable guarantees for the independent cascade model.
The Cost of Representation by Subset Repairs
Yuxi Liu*, Fangzhu Shen*, Kushagra Ghosh, Amir Gilad, Benny Kimelfeld, Sudeepa Roy
Proceedings of the VLDB Endowment (PVLDB), 2024 (presented at VLDB 2025). [pdf] [code]
Minimal-cost data repairs can quietly skew the representation of sub-populations, for example by deleting more tuples from one demographic group than another. We study the cost of representation, i.e., how many extra tuples must be deleted to satisfy both functional dependencies and representation constraints, and give complexity results, polynomial-time algorithms for special cases, and efficient heuristics in general.
Causal What-If and How-To Analysis Using HypeR Fangzhu Shen*, Kayvon Heravi*, Oscar Gomez, Sainyam Galhotra, Amir Gilad, Sudeepa Roy, Babak Salimi
International Conference on Data Engineering (ICDE), Demonstration Track, 2023. [pdf]
We demonstrate HypeR, an interactive system for expressing causal what-if and how-to queries in a SQL-like language and exploring their results visually.
Industry Experience
Uber · Ph.D. Software Engineer Intern
Mobility Simulation & Planning · Summers 2025 and 2026
2026 — Temporal Dependency Modeling
Improved the end-to-end learning framework by modeling temporal dependencies in marketplace forecasts.
2025 — End-to-End Marketplace Forecasting
Built an end-to-end learning framework integrating deep learning forecasts with a structural pricing model.
Education
Duke University, Durham, NC, USA
Ph.D. in Computer Science, January 2023 - Present
Expected graduation: January–May 2027
Advisor: Prof. Sudeepa Roy
GPA: 3.98/4.0
Duke University, Durham, NC, USA
M.S. in Economics and Computation, August 2021 - December 2022
Transitioned to the Ph.D. program
Central University of Finance and Economics, Beijing, China
B.A. in Finance, September 2016 - July 2020
Reviewer, VLDB 2026 Demonstration Track
Shadow PC Member, VLDB 2026
Awards
VLDB 2025 Travel Award
ICDE 2023 Travel Award
Duke Scholar Award, 2021-2022.
Academic Scholarships of Central University of Finance and Economics, 2017-2019.
