Vertical take-off and landing (VTOL) aircraft leverage the performance and efficiency of fixed-wing aircraft with the operational flexibility of rotary-wing aircraft. However, their complex dynamics and the inherent multi-mode of operation often make sufficiently robust controller design difficult, time consuming, and expensive. Transitioning between flight modes (e.g. vertical takeoff to cruise or cruise to landing) risks instability, oscillations, and high-G maneuvers. These challenges have been traditionally solved by relying on multi-modal PID design patterns and gain scheduling, necessitating the development of a different controller for each flight mode. Modern VTOL flight controller methods (e.g. Model Predictive Control) are often heavily model dependent and lack robustness to large uncertainties in the aircraft dynamic model. Such dependency eliminates any opportunity for flight controller to be transferable between different aircraft platforms without labor intensive practices. This paper presents simulation results of a deep learning based model-agnostic quad mode VTOL controller that is trained and optimized with the Proximal Policy Optimization (PPO) algorithm in simulation. Our controller is fully policy based rather than dynamic model dependent (e.g. LQR, H2, MPC, etc.) and thus is fully transferable between models without any retraining. Flight controller performance and robustness against perturbation is shown through a rigorous test suite that leverages data generated from real flight tests and the Dryden wind turbulence model.