This research introduces LLM4FL, an innovative multi-agent system that leverages large language models to identify software faults without requiring extensive training data.

• Combines graph-based retrieval with LLMs to understand code context
• Utilizes multiple specialized agents working together through a reflective process
• Achieves superior fault localization accuracy compared to traditional statistical methods
• Requires minimal computational resources while delivering enterprise-grade results

For software engineering teams, this approach offers a practical way to reduce debugging time and improve code quality with existing tools, potentially transforming how organizations approach quality assurance and maintenance.

A Multi-Agent Approach to Fault Localization via Graph-Based Retrieval and Reflexion