Novice software engineers often encounter difficulties adapting to collaborative workflows and practices required in professional software engineering environments. These challenges can lead to ineffective collaboration, unstructured contributions, and reduced project engagement. This thesis addresses these issues by developing an interactive dashboard designed to support novice engineers in understanding and improving their software engineering behavior.
The proposed dashboard extends the Hephaestus platform by aggregating collaborative activities from development repositories, including pull requests, issues, and code reviews. It visualizes the activities and detects bad practices, such as prolonged open pull requests, vague issue descriptions, and insufficient review comments. Leveraging large language models and project guidelines, the system automatically evaluates developer behavior and identifies violations of best practices. These findings are communicated through proactive notifications, offering users timely and actionable feedback. This approach promotes awareness and continuous learning.
The primary contribution of this work lies in its guideline-based bad practice detection with LLMs. Visualization and notification offer structured guidance for novices to improve collaboration. To evaluate the effectiveness of the detection, we collect quantitative usage metrics and assess the quality of detections. The results demonstrate the dashboard’s potential to enhance workflow adherence and collaboration among novice software engineers.