AI Application Offers Personalized, Tax-Aware Retail Portfolio Management

This paper describes a fully built and integration-tested application designed to bring personalized, tax-aware portfolio management to retail investors, a service typically reserved for institutional clients. Existing robo-advisors often rely on static, rule-based allocations, lacking the sophistication of institutional systems. The new application features a FastAPI backend and a web dashboard, allowing users to articulate investment goals in plain language, such as "steady growth but need to sell some shares next month." The system routes these goals to one of six investment mandates and generates live, broker-integrated portfolio recommendations using a three-phase reinforcement learning (RL) system. This RL system includes a self-supervised cross-asset encoder, a Mixture-of-Experts (MoE) allocation policy with a learned intent router, and a lightweight LoRA adapter for personalization based on individual brokerage behavior. While not yet live for end-users, the application has undergone end-to-end integration testing with a live brokerage API (Alpaca, paper-trading mode) and includes features like multi-user authentication and an auditable action-integrity chain. Preliminary 14-day walk-forward backtests show promising pre-deployment validation.