We develop a Physics-Informed Neural Network (PINN) code to solve the modified Teukolsky equation under realistic astrophysical conditions. The code embeds domain-specific physics—spin-weighted curvature perturbations, quasi-normal mode (QNM) boundary conditions, and attenuation dynamics—directly into the training loss function. Applied to data from the GW190521 event, the model accurately infers complex QNM frequencies (ω = 0.2917 − 0.0389i) and learns an attenuation coefficient α = 0.04096, corresponding to a 14.4% decay rate. The code demonstrates strong predictive performance, reducing mean squared error by 50.3% (MSE = 0.2537 vs. 0.5310) compared to Bayesian baselines, and achieving a positive R² score. It further reveals non-trivial r–t coupling and gravitational memory effects, which standard exponential decay models fail to capture. This PINN-based implementation establishes a computationally efficient and accurate tool for environmental modeling in gravitational wave astrophysics and offers a path forward for black hole spectroscopy beyond vacuum assumptions.