Predicting the vibrational and thermodynamic properties of materials at the pressures and temperatures of planetary interiors and of high-pressure synthesis requires a description of lattice dynamics beyond the harmonic approximation. The Temperature Dependent Effective Potential (TDEP) method provides that description by fitting effective force constants to forces sampled from finite-temperature configurations, but its cost is dominated by those force evaluations. Under density functional theory this restricts the approach to a small number of state points per material. This project replaces the density functional force evaluations with machine-learned interatomic potentials transfer-learned from foundation models (UPET and MACE) onto a small, targeted set of first-principles reference configurations for the system and pressure range of interest. Feasibility has been established in-house. A fine-tuned potential reproduces the first-principles energetics of a high-pressure hydride and drives a pressure-induced structural transformation that the unmodified foundation model does not describe. The result is predictive rather than reproductive: the reference configurations terminate part-way along the transformation path, while the potential located the product phase outside the sampled range. Fine-tuning is the only step requiring GPUs; reference-data generation, TDEP sampling and force-constant extraction run on CPU resources held elsewhere.