Hydrogen embrittlement (HE), i.e., the degradation of mechanical properties caused by hydrogen uptake, remains a major barrier to the proliferation of a hydrogen-based economy. HE is partly driven by the accumulation of hydrogen at microstructural defects, known as trap sites. These traps govern a material’s susceptibility to HE, making their characterisation essential for the design of HE-resistant alloys. Each trap type is defined by a binding energy and a trap density, collectively referred to as trapping parameters, which are commonly inferred from thermal desorption spectroscopy (TDS) experiments. The identification of these parameters from TDS data requires the use of simulation models and inverse calibration. To overcome the challenges encountered with conventional optimisation-based calibration approaches, this work introduces a machine-learning-based framework for parameter identification from TDS spectra. The proposed model demonstrates strong predictive performance when applied to three tempered martensitic steels of different compositions.