CV

Email zchen95@umd.edu · GitHub · Google Scholar · ORCID

Research interests

LLMs for forecasting and finance · Trustworthy and secure LLM agents · LLM evaluation and LLM-as-a-judge · Data-centric AI and computational social science · Multilingual NLP

Education

University of Maryland, College Park2024–present

Ph.D. in Information Science · GPA 4.0/4.0 · Dean's Fellowship
Coursework: Natural Language Processing, Generative AI Agents, Long-Context Language Models, Multilingual LLMs, Commonsense Reasoning and NLU, Advanced Machine Learning, Constructivist Grounded Theory

University of Maryland, College Park2019–2020

Master of Public Administration · GPA 3.94/4.0

Guangdong University of Foreign Studies2017–2020

M.A. in Translation Studies · GPA 92/100

Guangdong University of Technology2013–2017

B.S. in Computer Science; B.A. in English for Science and Technology · GPA 3.99/4.0

Experience

International Monetary Fund, Monetary and Capital Markets Department2021–2024

Research Analyst, Central Bank Operations Division

Co-developed CoFiE-NN, an LSTM with Cornish–Fisher expansions for Value-at-Risk forecasting that outperformed EGARCH-t and EVT across 22 currencies; built the backtesting framework.
Built a reserve-demand model (WP/23/179) used in the IMF's 2024 U.S. Article IV report to estimate the Fed's reserve needs (US$2.1–3.2 trillion).
Developed and taught Python/R/MATLAB liquidity-forecasting tools delivered to 15+ central banks; first-authored the IMF handbook chapter on their statistical methods.
Joined 25+ technical assistance and FSAP missions, including stress testing of liquidity, solvency, and interconnectedness for Saudi Arabia and Panama.
Automated extraction of Bank of Japan securities-operations data from PDFs and the web.

United Nations Development Programme, Bureau for Policy and Programme Support2020–2021

Policy Analyst Intern

Compiled and analyzed economic and sovereign debt data for 136 countries, producing simulations, charts, and tables for policy research.
Built a global COVID-19 vaccine administration and cost dataset and estimated country-level spending under multiple scenarios.
Contributed analysis and policy interpretation to research papers and policy briefs.

Research projects

See the Research page for details.

TimeJudge: An Agentic Framework for Judgmental Forecasting (in progress)
Robust Attention Shield for Prompt Injection (in progress)
LLM Deductive Coding with Controlled Synthetic Data (in progress)
Huber DPO: A Robust Framework for LLM Alignment (course project, 2025)
Towards Pragmatic Realism in Synthetic Code-Switched Data (course project)
Containing Catastrophic Forgetting in Code-Switched Multitask Continual Fine-tuning (course project)
Lost in Translation: Stress-Testing Greater Chinese Cultural and Pragmatic Nuance (course project)
Demojification of Online Social Biases (course project)

Publications

Working Papers

Forecasting Tail Risk via Neural Networks with Asymptotic Expansions

Yuji Sakurai, Zhuohui Chen

IMF Working Paper WP/24/99, 2024

IMF page PDF

Modeling the Reserve Demand to Facilitate Central Bank Operations

Zhuohui Chen, Nikolaos Kourentzes, Romain Veyrune

IMF Working Paper WP/23/179, 2023

Presented at the International Symposium on Forecasting, 2023. The model was used in Box 7 of the IMF's 2024 U.S. Article IV report to estimate the Fed's long-run reserve needs.

IMF page eLibrary