Research Projects
I work in the SOAR Lab, where we advance software engineering with large language models and AI-driven methods. Below are some of my recent projects and publications:
How Do LLMs Read Bug Reports? Attention Analysis of LLMs
ASE 2026
What it is. The first empirical study of how large language models allocate attention over bug rerports while repairing real-world software bugs, comparing model attention with developer reasoning.
Why it matters. Shows that successful repairs rely on distributed attention to diagnostic information, while attention misallocation is a key factor behind repair failures, providing insights for more reliable and interpretable AI systems.
📄 Paper
A Multivocal Literature Review on Technical Debt in LLM-Assisted Software Development
TOSEM Journal 2026
What it is. A multivocal literature review of 104 research papers and industry sources examining how LLM-assisted development introduces and amplifies technical debt.
Why it matters. Identifies emerging forms of technical debt unique to AI-assisted coding, highlights gaps in current evaluation methods, and outlines future directions for building more sustainable AI-powered software development.
📄 Paper
Hierarchical Knowledge Injection for LLM-based Program Repair
ASE 2025
What it is. A layered retrieval pipeline (bug → repository → project) that supplies LLMs with the right level of context for the purpose of automated program repair.
Why it matters. Improves correctness across different bug types and shows which levels of context LLMs actually need to generate effective repairs.
Towards Detecting Prompt Knowledge Gaps for Improved LLM-guided Issue Resolution
MSR 2025
What it is. A tool (browser extension) that detects missing context, specifications, and unclear prompts in developer–LLM conversations that could hinder bug fixing.
Why it matters. Helps developers craft better prompts so LLMs can return correct fixes faster.
What Characteristics Make ChatGPT Effective for Software Issue Resolution?
Empirical Software Engineering Journal 2025
What it is. Analyzes a large corpus of developer–ChatGPT conversations to understand why developers seek help for software bugs and when these interactions lead to successful resolutions.
Why it matters. Provides empirical insights into the qualities of effective developer–AI conversations for software problem solving.
A Comprehensive Annotated Dataset of Locked GitHub Issue Threads
MSR 2024
What it is. Introduces a large annotated dataset of GitHub issue threads that were locked due to incivility, capturing a range of uncivil interactions in open-source communities.
Why it matters. Offers a rich resource for studying online incivility and designing interventions to improve open-source communities.
Exploring Moral Principles Exhibited in OSS Communications
FSE 2023, NLBSE 2025
What it is. An NLP study of open-source discussions uncovering moral principles behind developer arguments, paired with an ML-based tool to flag toxic conversations.
Why it matters. Shows how moral reasoning can improve toxic conversation detection in open-source projects.