Multi-Agent LLM Collaboration for Adaptive Code Review, Debugging, and Security Analysis

Automated code review systems have improved software development, yet many lack contextual awareness, leading to redundant feedback and limited adaptability to user-specific coding styles. This paper presents a multi-agent AI-driven framework that leverages FAISS-based memory to improve review efficiency, personalization, and collaboration. The system integrates code review, bug detection, and security analysis agents, operating independently and interactively to analyze and refine code submissions. A feedback mechanism enables users to influence future AI suggestions, ensuring that recommendations evolve based on individual preferences and past interactions. Through structured experiments, we evaluated the impact of FAISS memory on the reduction of redundant feedback, evaluated the effectiveness of collaborative agent execution, and measured the system’s ability to adapt to coding patterns over time. The results indicate that the incorporation of FAISS significantly improves AI-assisted development by minimizing unnecessary repetitions while maintaining essential corrective feedback. This research demonstrates the potential of adaptive AI systems in software engineering, contributing to more intelligent, context-aware, and efficient code review methodologies.