CV
Zehua Zhang
Third-Year PhD Student in Computer Science
Summary
Third-year PhD student at Arizona State University researching natural language processing, LLM post-training, multi-agent systems, and AI for software security.
Education
- PhD in Computer ScienceExpected Spring 2028Arizona State UniversityGPA: 3.67 / 4.0
- M.S. in AI Engineering - Information SecurityMay 2024Carnegie Mellon UniversityGPA: 3.7 / 4.0
- B.A. in Computer Science and EconomicsMay 2022Boston CollegeGPA: 3.65 / 4.0
Work Experience
- Research InternMay 2026 - Aug 2026Samsung Research AmericaConducted research on indirect prompt injection vulnerabilities in LLM-based agentic systems.
- Developed a gym-like environment for scalable training-data sampling and generation.
- Trained LLMs using supervised fine-tuning and reinforcement learning methods, including DPO and GRPO.
- Achieved state-of-the-art performance on AgentDojo and AgentDyn.
- Algorithm Development InternJune 2021 - Sept 2021Sohu Inc.Designed and deployed a real-time feed deduplication system for a social media application.
- Used Kafka, Spark, simhash, and Bloom filters for large-scale multimedia deduplication.
- Reduced repetitive text exposure by approximately 90% and duplicate video content by approximately 30%.
Skills
Programming
- Python
- C
- Scala
- Java
- R
- Swift
- JavaScript
- HTML
- CSS
ML/DL Frameworks
- PyTorch
- TensorFlow
- PyG
- Verl
- vLLM
Data and Distributed Systems
- Spark
- Kafka
- Kubernetes
Cloud and Databases
- MySQL
- MongoDB
- Hive
- HBase
- GCP
- AWS
Software Development
- React
- iOS
- Docker
- MCP
Publications
- CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents2026
- Build-Bench: Benchmarking LLM Agents on Compiling Real-World Open-Source Software2026
- When Competency in Reasoning Opens the Door to Vulnerability: Jailbreaking LLMs via Novel Complex Ciphers2025NeurIPS 2025 Reliable ML from Unreliable Data Workshop
- Masked Pretraining Strategy for Neural Potentials2024
- Enhancing LLM Factual Accuracy with RAG to Counter Hallucinations: A Case Study on Domain-Specific Queries in Private Knowledge-Bases2024
- One first-authored paper submitted to NDSS 20272026
Teaching
- Information Assurance (CSE 365)2026Arizona State UniversityRole: Graduate Teaching AssistantGrading, office hours, and student support.
- Natural Language Processing (CSE 576)2024Arizona State UniversityRole: Guest LecturerDelivered a lecture about language model-based agents.
- Randomness and Computation (CSCI2244)2021Boston CollegeRole: Undergraduate Teaching AssistantGrading, office hours, and student support.
- Web Application Development (CSCI2254)2022Boston CollegeRole: Undergraduate Teaching AssistantGrading, office hours, and student support.
Interests
- ResearchNatural Language Processing, LLM Post-Training, Multi-Agent Systems, AI for Software Security