This work introduces CODECOM, an intelligent code completion technique that leverages a large language model trained using a unified multi-task learning framework to provide accurate and context-aware code predictions.

This work introduces CODECOM, a deep learning-based code completion technique that integrates a large language model with program analysis. By leveraging source code tokens, abstract syntax trees, and program dependencies, CODECOM achieves state-of-the-art performance and improves developer productivity.

This work presents DEEPKNOWCODE, an automated approach for generating natural language summaries of source code using a large language model trained with crowdsourced knowledge.

DEEPKNOWCODE leverages crowdsourced knowledge from GitHub and Stack Overflow to produce context-aware code summaries that explain program behavior, implementation rationale, and usage guidelines, significantly improving program comprehension.

This work introduces SYNCode, a collaborative framework that combines human expertise and large language models to improve data annotation quality for Stack Overflow datasets.

SYNCode enables human–LLM collaboration through a multi-stage pipeline integrating semantic modeling, code-aware embeddings, and iterative human validation, resulting in higher-quality annotations with reduced manual effort.

This work presents SSDTutor, an intelligent tutoring system that provides automated feedback to support secure software development education.

SSDTutor detects security vulnerabilities and insecure coding practices in student-written code and delivers personalized feedback, improving learners’ understanding of secure software engineering principles.

This work proposes RIT, a technique for improving regression testing efficiency after refactoring activities.

RIT combines AST-based refactoring detection, program slicing, and change impact analysis to localize failure-inducing edits, reducing testing and debugging effort during software evolution.

This work introduces PRI, a static-analysis-based approach for inspecting clone refactorings in evolving software systems.

PRI summarizes clone refactorings and detects inconsistent or incomplete changes using AST-based analysis, improving software quality and maintainability.

This work presents an automated approach to inspect manual refactoring edits that are prone to subtle errors.

The approach combines change analysis, structural dependency analysis, and refactoring-aware inspection to improve refactoring correctness in evolving software systems.

This work introduces CRITICS, an interactive code review technique for inspecting systematic and cross-cutting changes.

CRITICS generalizes edits into AST-based templates to detect missing or inconsistent updates, significantly improving code review accuracy and efficiency.

This work introduces metadata invariants, a declarative abstraction for capturing implicit conventions between source code and metadata.

The approach defines the Metadata Invariants Language (MIL) and inference algorithms that enable continuous checking of metadata correctness during software evolution.

This work proposes a framework that enables source-level programming tools to remain effective in the presence of transparent bytecode transformations.

By introducing Structural Enhancement Rules, the approach allows editors and debuggers to reflect runtime behavior accurately while preserving source-level abstractions, improving program comprehension.