Recent advances in GPU-accelerated computational fluid dynamics (CFD) and machine learning (ML) are enabling higher-fidelity digital engineering workflows by reducing turnaround time for large-scale simulations and accelerating data-driven surrogate development. This work presents an end-to-end framework for generating ML surrogates using GPU-driven overset simulations within the HPCMP CREATE-AV Helios environment, leveraging RAPIDUS (near-body unstructured solver), ORCHARD (off-body Cartesian solver), and PUNDITG (overset connectivity) executed in a performance-portable GPU mode. Using the Hover Validation and Acoustic Baseline (HVAB) rotor as a representative rotorcraft configuration, high-fidelity simulations are used to train deep neural network (DNN) surrogates for isolated rotor torque across a bounded operational domain in rotor collective (4-10 deg) and non-dimensional climb condition. A key feature is an adaptive training strategy that combines cross-validation with Voronoi-based candidate selection to identify regions of heightened sensitivity and add samples where they most improve model robustness. For the HVAB domain examined, a compact dataset was shown to be sufficient: an initial uniform design was refined through adaptive cycles to a final distribution (order 10s of points) while improving agreement against independent validation cuts. Once trained, the surrogate enables rapid batch prediction (seconds) suitable for design studies and trade-space exploration. Timing analyses demonstrate the practical impact of GPU-driven truth generation: engineering-fidelity cases converge in under an hour on multiple A100 GPUs, enabling completion of a high-fidelity surrogate within roughly a day. Comparable CPU-based workflows require substantially longer and are more sensitive to resource availability. Ongoing extensions for WCCM 2026 will include workflow streamlining and single-source multi-fidelity formulations that combine low- and high-fidelity predictions to further reduce lead time while preserving accuracy in nonlinear regimes.