Abstract:
Problem: Detecting complex software bugs, such as logical errors, performance inefficiencies,
and concurrency issues, remains a significant challenge for development teams. Traditional static
analysis tools struggle to capture contextual dependencies, while dynamic analysis methods often
require extensive computational resources. These limitations delay bug detection, increasing
development costs and technical debt. This research aims to enhance bug detection by leveraging
AI-driven techniques within an integrated development environment (IDE), specifically
improving generative AI’s ability to identify and address complex code dependencies.
Methodology: This project introduces an AI-powered bug detection plugin designed for
real-time integration with an IDE. The system follows a multi-level approach, beginning with
syntax error detection and extending to logical error analysis. A hybrid backend agent
orchestrator coordinates the execution of core analysis tools, including a Syntax Error Checking
Tool, a Refactoring and Fix Suggestion Tool, and a Logical Error Detection Module. The plugin
leverages a fine-tuned LLaMA 3.2 1B model trained on datasets such as Defects4J and
QuixBugs to improve accuracy in detecting Java bugs. The implementation ensures a
lightweight, configurable workflow, allowing developers to tailor error detection based on
project requirements.
Results: The fine-tuned model achieved an accuracy of 85.81% in bug detection, surpassing the
base model’s 76.66% performance. Testing on large-scale Java projects confirmed the system’s
ability to analyze complex codebases with minimal latency. The plugin effectively detects and
suggests fixes for syntax and logical errors, enhancing developer productivity by providing
contextual bug resolutions within the IDE. Future enhancements will focus on expanding
language support, improving detection for performance-related issues, and refining the AI model
for broader applicability in software debugging.
