XDipin Khati
About Me
I am a PhD candidate in Computer Science at the College of William & Mary, working under the supervision of Dr. Denys Poshyvanyk. My research focuses on applying Deep Learning and Large Language Models (LLMs) to Software Engineering tasks, with a particular emphasis on making these models more trustworthy, interpretable, and reliable for real-world software development.
Research Interests
My research interests span several key areas:
- Interpretable AI for Code: Developing methods to understand and explain how LLMs generate and reason about source code
- Trustworthy Machine Learning: Building techniques to assess and improve the reliability of AI systems in software engineering contexts
- Causal Inference in Software Engineering: Applying causal analysis to understand the factors that influence model performance and reduce confounding bias
- Syntax-Aware Code Generation: Leveraging Abstract Syntax Trees (ASTs) and programming language semantics to improve code generation quality
- Prompt Engineering and Evaluation: Creating better benchmarks and evaluation strategies for LLMs applied to software tasks
Current Work
Currently, I am working on developing interpretability methods for Large Language Models when applied to code-related tasks. My recent work includes:
- Mapping the Terrain (TOSEM)
- ASTrust (Under Review)
- Global Code-Based Explanations (Under Review)
- Galeras (ICSME 2023)
- Code Smells in LLM-Generated Code (ICSE 2026)
Mapping the Terrain (TOSEM): A comprehensive survey and perspective on the use of LLMs in software engineering, mapping the current landscape of research and applications.
ASTrust (Under Review): A novel interpretability framework that generates syntax-grounded explanations by mapping model confidence to Abstract Syntax Tree structures, enabling developers to understand model predictions at both local and global levels.
Global Code-Based Explanations (Under Review): An approach to explain LLMs for code using global code-based explanations, providing comprehensive understanding of model behavior across different code contexts.
Galeras (ICSME 2023): A benchmarking strategy for evaluating LLMs in software engineering tasks using causal inference, helping researchers quantify treatment effects and reduce confounding bias in model evaluation.
Code Smells in LLM-Generated Code (ICSE 2026): A causal perspective on measuring, explaining, and mitigating code smells in LLM-generated code, introducing the Propensity Smelly Score (PSC) to assess structural quality.
Looking for Opportunities
I am actively seeking summer research internships for 2025 where I can contribute to cutting-edge research in:
- Machine Learning for Software Engineering
- Large Language Models and Code Generation
- AI Interpretability and Trustworthiness
- Software Engineering Tools and Automation
If you’re interested in collaborating or have internship opportunities, please feel free to reach out!
Recent News
Quick Links
Contact
I’m always interested in discussing research opportunities, collaborations, or internship positions. Please feel free to reach out via email (see sidebar) or through the contact information in my CV.