<p>Gravity inversion is a fundamental method for modelling subsurface density distributions using measured surface gravitational acceleration anomalies. The Adaptive Moment Estimation (ADAM) algorithm is applied to two-dimensional gravity inversion in this study, introducing the ADAMGRAV approach to enhance convergence stability and accuracy. Synthetic experiments involving horizontal, combined horizontal-inclined prisms, and complex Horst-Graben models show that the algorithm converges rapidly without oscillations. It accurately reconstructs both density contrasts and structural geometries for the prism models, achieving high correlation coefficients of 0.90 and 0.81. For the highly complex Horst-Graben structure, the algorithm successfully identifies main structural boundaries despite a lower correlation coefficient of 0.28, demonstrating its realistic capability in handling challenging non-linear inverse problems. The approach is further validated using residual Bouguer anomaly data from the Lee-Allen geothermal prospect in Nevada, USA. Inversion results are integrated with existing three-dimensional Magnetotelluric (MT) modelling and geologic interpretations from the BRIDGE project. The ADAMGRAV inversion delineates a basement rock body with a high positive density contrast (greater than 0.35 g/cm³) that acts as a low-permeability structural barrier, bounded by the northwest-striking Russell Pass strike-slip fault. Additionally, an adjacent negative density contrast anomaly (less than -0.25 g/cm³) is associated with a hydrothermal upflow pathway and a shallow outflow zone. The ADAMGRAV inversion achieves low final misfits of 0.38, 0.42, and 1.18 in synthetic tests, and 0.34 in the field data application. These results demonstrate that the ADAMGRAV inversion produces geologically plausible structural models, establishing it as a robust and stable alternative to traditional least-squares solvers for geothermal and geophysical exploration.</p>