Mohab M. Metwally · FinAI — Preprint · 2026
Robo-advisory platforms now manage trillions of dollars in assets, yet they typically operate as black boxes, exclude emerging asset classes such as cryptocurrencies, and compute static allocations that cannot adapt trading timing to market dynamics. This paper presents FinAI, an open-source platform that unifies Modern Portfolio Theory (MPT) for strategic asset allocation with deep Reinforcement Learning (RL) for tactical trade execution within a single transparent system. On a large-cap equity universe the maximum-Sharpe strategy attains a Sharpe ratio of 1.45, significantly exceeding an equal-weight baseline (1.28) with an 18% reduction in maximum drawdown (paired t-test, p = 0.005), and a PPO agent achieves a Sharpe ratio of 1.12 out-of-sample against 0.88 for buy-and-hold, with 37% lower drawdown. The results indicate that combining classical optimisation with modern deep RL in a transparent, professionally validated framework yields measurable, statistically significant improvements over standard baselines.