CV

Download full CV (PDF)

Education

• Ph.D. in Computer Science, North Carolina State University, Aug. 2023 - Present
  • Advisor: Dr. Wesley K. G. Assunção
  • Research Interests: LLMs, Software Engineering, HCI, AI4AI, AI4SE, Agents
• M.S. in Computer Science, University of Florida, Aug. 2020 - Dec. 2022
• B.E. in Software Engineering, Xidian University, Aug. 2015 - Jun. 2019

Work experience

• May 2026 - Aug. 2026: Core-ML Ph.D. SWE Intern
  • Google LLC, Mountain View, CA
  • Host: Ann Yan
  • Architected an Autonomous Agent Platform: Built an end-to-end LLM agent in Python to automate migrating PyTorch models onto Google Cloud TPUs; implemented a 20+ tool MCP server covering AST analysis, automated dependency resolution, and vector RAG.
  • Engineered Closed-Loop Diagnostics & Verification: Developed a two-stage self-healing pipeline using dry-run profiling to detect XLA compiler CPU fallbacks before full-scale training; delivered a production-grade platform across 35+ modules.
• Jan. 2023 - Aug. 2023: Software Engineer, Full Time
  • PAX Technology Inc., Jacksonville, FL
  • Mentor: Yurii Maidaniuk
  • Contributed to a return merchandise authorization system managing repairs and tracking equipment for over 3,400 client companies in North America, from initial documentation to final deployment.
  • Participated in developing a warehouse management system managing 1.4 million devices.
  • Built the counting tax function of a Key Injection purchase system using Spring Boot, generating approximately $840,000 in annual revenue.
  • Contributed to the CyberLab Function, a government-approved robotics kit used by global testing teams to evaluate financially sensitive features such as PIN entry.
• Mar. 2021 - Jun. 2021: Software Engineering Intern
  • IBM, Shanghai, CN
  • Developed and deployed a monitoring function for the Development Bank of Singapore’s bank card system, enhancing security across all branches in Asia.
  • Used IBM WebSphere to control and deploy different versions of the Bank Card System, helping manage $134.5 billion in deposits for DBS Bank.
  • Wrote a product requirement document defining raw data formats to support IBM’s Singapore and Hong Kong teams in developing Blacklist/Whitelist functions.

Selected projects

• Evaluating Language Models on Cross-Language Code Functional Equivalence (Aug. 2025 – May 2026), Google Cloud, FUNCAP, CNPq funded project
  • Built PolyHuman, the first dataset for evaluating cross-language code functional equivalence on human-written code (C++/Java/Python, 5,035 problems from CodeContests, 15 sub-tasks).
  • Systematic evaluation of 9 open- and closed-source models shows accuracy drops from 0.998 on synthetic code to 0.84 on human-written code.
  • Manual analysis of 81 misjudged cases produced a three-tier failure hierarchy: Knowledge Failures (21%), Abstraction-Level Reasoning Failures (53%), and dataset-labeling issues (19%).
• Understanding Reviewer Concerns on Test Code in Pull Requests (Dec. 2024 – Mar. 2025), NSF funded project in the WISER lab
  • First analysis of GitHub Actions’ influence on test code review, leveraging 213,511 pull requests and 600,905 review comments from six large-scale GitHub projects.
  • Found that adoption of GitHub Actions significantly reduced reviewer and developer attention to test code.
• Software-hardware Co-design Differential Privacy (Aug. 2023 – May 2024), NSF funded project in the Network Lab
  • First to leverage hardware transistor degradation under voltage drop as a noise source for Local Differential Privacy.
  • Proposed SRAM_DP, a hardware-software co-design for differential privacy on semi-structured data. Published at ISVLSI 2024.

Skills

• Agent building, RAG, XLA, Java, Python, C++, R, Matlab, SQL, J2EE, ASP.NET, AI4SE, LLMs, LaTeX, Git
• Frameworks & Libraries: Spring Boot, MyBatis, PyTorch, scikit-learn, Ollama, Pandas, HuggingFace

Publications

From GPU to TPU with Coding Agents: What Works, What Fails, and What It Costs?

Hui Sun, Anderson Uchôa, and Wesley K. G. Assunção. "From GPU to TPU with Coding Agents: What Works, What Fails, and What It Costs?" arXiv:2610.34422, 2026. [Under Review at ICSE-SEIP 2027]

Same Workflow, Different Representation: Effects on LLM-Based GitHub Actions Smell Detection

"Same Workflow, Different Representation: Effects on LLM-Based GitHub Actions Smell Detection." [Under Review at SANER 2027]

NetSight: Joint Spatio-Temporal Dependency Modeling for Network Traffic Prediction

Jinming Xing, Guoheng Sun, Hui Sun, Linchao Pan, Shakir Mahmood, Xuanhao Luo, and Muhammad Shahzad. "NetSight: Joint Spatio-Temporal Dependency Modeling for Network Traffic Prediction." In Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM), 2026.

Beyond Smell Detection: Construct-Aware Evaluation of Agentic LLM Pipelines for GitHub Actions Workflow Refactoring

Anderson Uchôa, Jhefferson Souza, Emanuel Ávila, Gabriel Silva, Kevin Lira, Nelson Araújo, Arthur Cordeiro, Hui Sun, Rohit Gheyi, and Wesley K. G. Assunção. "Beyond Smell Detection: Construct-Aware Evaluation of Agentic LLM Pipelines for GitHub Actions Workflow Refactoring." [Under Review at ICSE 2027]

Evaluating Language Models on Cross-Language Code Functional Equivalence

Hui Sun, Anderson Uchôa, Rohit Gheyi, and Wesley K. G. Assunção. (2026). "Evaluating Language Models on Cross-Language Code Functional Equivalence." International Symposium on Empirical Software Engineering and Measurement (ESEM 2026). Google Cloud Research Award Supported.

Test Code Review in the Era of GitHub Actions: A Replication Study

Hui Sun, Yinan Wu, Wesley K. G. Assunção, and Kathryn T. Stolee. "Test Code Review in the Era of GitHub Actions: A Replication Study." [Under Double-Blind Review]

Low-power and Computing-free Privacy Design for IoT Systems

Hui Sun, Kyle Mooney, Mario Renteria-Pinon, Tingxiang Ji, Hritom Das, Na Gong, and Jianqing Liu. (2024). "Low-power and Computing-free Privacy Design for IoT Systems." IEEE Computer Society Annual Symposium on VLSI (ISVLSI).