Curriculum Vitae
Education
-
Columbia University, New York, US
PhD in Computer Science M.S. in Computer Science (Feb 2026)
B.A. in Computer Science (May 2024, GPA: 3.73)
Honors: Academic Scholarship, Dean’s List -
Institut d’études Politiques (IEP) de Paris, Paris, FR
B.A. in Economics (May 2024, GPA: 4.00)
Experience
- Columbia University, Research Assistant — New York, US
Jan 2024 – May 2025
Formal Verification, Program Synthesis, and AI.- Applied synthesis methods to increase safety and predictability of LLM-generated code.
- Used natural language to generate complex arbiters through LLMs.
- Columbia University, Research Assistant — New York, US
Aug 2023 – May 2025
Deep Reinforcement Learning for Finance.- Lead RA on SIRS/STAR-funded project improving LLM understanding of code and tabular data.
- Built GPU-parallel RL frameworks for financial trading.
- Managed team of six researchers and coordinated ICAIF’24 financial AI competition.
- IBM Research, Research Assistant — New York, US
Oct 2024 – Dec 2024
Co-design of resistive processing units (RPU) for analog in-memory compute systems.- Improving energy-efficient AI through neural architecture search (NAS) using IBM AIHWKit.
- Developing robust analog deep learning architectures accounting for nonlinearities.
- Columbia University, Course Assistant — Data Structures and Algorithms
Sep 2023 – May 2024- Assisted in course delivery for 400+ students; ran review sessions and recitations.
- Baqend GmbH, Full-Stack Developer Intern — Hamburg, DE
Aug 2020 – Sep 2021- Built proof-of-concept full-stack apps for five enterprise clients.
- Supported production systems used by millions of users weekly.
Publications
- Holzer, N., Wang, K., Xiao, K., & Yanglet, X.-Y. L. ( 2025 ). Revisiting ensemble methods for stock trading and crypto trading tasks at ACM ICAIF FinRL Contest 2023-2024 . ArXiv Preprint ArXiv:2501.10709 .
- Murphy, W., Holzer, N., Qiao, F., Cui, L., Rothkopf, R., Koenig, N., & Santolucito, M. ( 2024 ). Combining llm code generation with formal specifications and reactive program synthesis . ArXiv Preprint ArXiv:2410.19736 .