Software Engineering Research

Cerebra aligns implicit knowledge between users and LLMs during SQL authoring by retrieving context from historical scripts and supporting iterative refinement.

Study comparing semantics-first, block-based, and textual programming for secondary students finds that making program state visible during coding yields significantly better task performance.

Controlled experiment examining how GUI versus API interfaces affect expert performance in symbolic execution for software vulnerability discovery and tool usability.

Spectral graph metrics derived from NLP-extracted requirement networks predict integration effort with 95% correlation, outperforming traditional structural complexity measures for early-stage.

Web-based AI system integrating code execution, error detection, and AI explanations across multiple languages to improve programmer comprehension and debugging efficiency

Qualitative and quantitative study examining how web developers adopt AI tools, their efficiency gains, and persistent concerns about code quality and security vulnerabilities.

Systematic review of 189 papers examining Large Language Models for Automated Program Repair from 2020-2025, analyzing LLM architectures, deployment strategies, and applications in bug repair and.

Review of machine learning techniques for software fault prediction from 2023-2025, examining algorithms, datasets, evaluation methods, and challenges in predictive modeling for software reliability.

Study examines how major AI companies implement guardrails as sociotechnical control mechanisms, revealing how code and language jointly regulate discourse in large language models.

Systematic research agenda for migration-based software maintenance automation, establishing a four-stage lifecycle model for transferring knowledge and solutions across software systems.

FLAT uses context-free grammars as type definitions to distinguish semantically different string encodings, enabling type-safe handling of URLs, file paths, and other structured data.

Study integrating LLMs with interactive visual aids for code comprehension, evaluating improved understanding and user trust through multimodal interaction design.

Lightweight on-device vision-language model generating HTML from design mockups without cloud infrastructure, supporting private prototyping with 235M parameters achieving competitive results.

Learn how retrieval augmentation and structural priors enhance parameter-efficient code representations while using only 5% of standard fine-tuning parameters.

Modular verification in dependent type theory using abstraction functions as types, with support for cost analysis and behavior verification while maintaining privacy guarantees.