CV
This is a description of the page. You can modify it in '_pages/cv.md'. You can also change or remove the top pdf download button.
Contact Information
| Name | Xiaoyan Guo (Flora) |
| Professional Title | Student |
| floraa@mail.ustc.edu.cn |
Professional Summary
Research interests include AI-assisted software engineering, large language models, software testing, and program analysis. Incoming PhD student at the University of Virginia.
Experience
-
2024 - 2026 Hefei, Anhui, China
Bachelor Research Assistant
University of Science and Technology of China
Explored research topics in artificial intelligence and large language models, later focusing on software engineering and LLM-assisted software development.
-
2025 - 2025 Hong Kong
Research Assistant
The Hong Kong University of Science and Technology
Research on AI-driven multi-agent systems for adaptive code completion and AI-assisted software development.
Education
Publications
-
Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization
ICML 2026
Proposed a difficulty-aware group normalization method to improve multimodal LLM reasoning in reinforcement learning settings.
-
2025 Multi-Agent Systems for Dataset Adaptation in Software Engineering Capabilities, Limitations, and Future Directions
arXiv
Studied the effectiveness and limitations of multi-agent systems for adapting code completion methods to unseen software engineering datasets.
Projects
-
LLM-Assisted C-to-Rust Concurrent Program Migration
Research on leveraging large language models and verification tools to automate the migration of concurrent C programs into safe and idiomatic Rust code.
- Designed an iterative multi-stage migration pipeline combining LLM-based translation with automated verification and repair.
- Evaluated concurrency safety, functional equivalence, and Rust idiomaticity using tools such as Loom and Clippy.
- Conducted empirical studies on benchmark concurrent programs to analyze migration correctness and robustness.
-
AI-Driven Multi-Agent Frameworks for Adaptive Code Completion
Research on adaptive multi-agent systems for AI-assisted code completion and automated prompt engineering.
- Evaluated cross-project adaptation and robustness of multi-agent code completion systems.
- Developed prompt-fixing strategies and automated evaluation workflows.
- Built a VSCode extension for Python-based GitHub Copilot interaction.
-
Enhancing MLLMs Reasoning via Difficulty-Aware Group Normalization
Research on improving multimodal large language model reasoning with difficulty-aware reinforcement learning strategies.
- Designed difficulty-aware regrouping methods for RLVR training.
- Evaluated performance across multimodal reasoning benchmarks.
- Achieved consistent improvements over GRPO/DAPO baselines.
-
Knowledge-Enhanced Agents Based on RAG and Web Search in LLM
Developed knowledge-grounded LLM agents integrating retrieval-augmented generation and real-time web search.
- Led implementation of the WebSearch subsystem with dual-agent architecture.
- Developed plugins for Google Search, Google Scholar, and arXiv integration.
- Enhanced retrieval and planning modules for grounded generation.