Pishehvar's RL app brings tax-aware, goal-based investing to retail traders

Pishehvar's application targets a gap between static rule-based robo-advisors and institutional-grade systems. Built on FastAPI with a web dashboard, it lets retail investors type natural-language goals—like "I want steady growth but need to sell shares next month for a down payment"—then routes the request to one of six investment mandates. The core is a three-phase RL stack: a self-supervised cross-asset encoder, a Mixture-of-Experts (MoE) allocation policy with a learned intent router, and a lightweight LoRA adapter that personalizes recommendations from an individual's revealed brokerage behavior without retraining the shared model. The system is integration-tested end-to-end against Alpaca's paper-trading API, including multi-user authentication, a trust-first preview-before-apply confirmation flow, daily email digests, and an auditable action-integrity chain.

While functionally complete, the app has not yet opened to real end-users, and the author honestly labels it as a pre-deployment application. Validation comes from 14-day walk-forward backtests with bootstrapped confidence intervals, presented as preliminary rather than production performance. The paper also shares practical engineering lessons—silently-inactive integration paths, hanging third-party API calls, and the value of end-to-end empirical verification over trusting checkpoint metadata—that generalize to other applied RL systems on live external data. The work demonstrates a feasible path to democratizing tax-aware, personalized portfolio management for retail investors.