A BIASAWARE MULTI-AGENT LARGE LANGUAGE MODEL FRAMEWORK FOR SAFE AND EXPLAINABLE RETAIL INVESTMENT ADVISORY

Large Language Models (LLMs) have recently become widely used by private investors seeking personalized financial guidance. Yet emerging research indicates that such models systematically amplify several investment risks, including excessive geographic concentration, sector clustering, trend chasing, elevated active management exposure, and increased total expense ratios. At the same time, Multi-Agent LLM architectures have demonstrated notable improvements in quantitative financial analytics by integrating specialized reasoning components, structured workflows, and dynamic code execution. This paper combines these two streams of insight by introducing a novel Bias-Aware Multi-Agent LLM Framework designed specifically for retail investment advisory. The proposed system incorporates risk auditing, bias detection, regulatorystyle constraint enforcement, and user-centered explanation mechanisms into a Multi-Agent foundation. Experimental evaluations show that architecture substantially reduces key investment risks while preserving clarity, interpretability, and rigorous analytical performance. The work aims to move toward safer, more transparent, and more responsible AI systems for consumer-facing financial applications.