Laser-based direct energy deposition (L-DED) is characterized by an inherent heat accumulation, resulting in a dynamic deposition process that presents challenges in monitoring and qualification of the printed part. This research proposes a novel approach to optimize the buildability of L-DED components by varying the laser power supply to control the amount of heat introduced during the deposition process ensuring a uniform melting and fusion of metal powder. The work focuses on virtual real-time monitoring of the L-DED process by integrating a machine learning model within the computational domain to maintain a consistent melt pool formation. An artificial neural network (ANN) was developed through hyperparameter tuning to create an optimized architecture that captures the complexity of the L-DED process. Data for the model was sourced from simulations carried out using FEMUSS, an in-house software package developed at CIMNE for the High-Performance Computing (HPC) simulations of complex multi-physics problems (Moreira et al., 2024). The input features of the ANN include deposition temperatures and melt pool dimensions while the prediction of the model is the optimal power supply. A multiple linear regression model was employed to fit the input data whereby loss curves were generated to evaluate the training process (Herzog et al., 2024). Moreover, the predictive capabilities of the model were validated through testing on a separate unseen dataset and a high coefficient of determination was observed on the prediction versus actual value plots indicating robust generalization abilities. In conclusion, the multiple linear regression model effectively captures the complexity of the L-DED deposition process, demonstrating a strong correlation between selected input features and the output label. The optimized power profile generated by the virtual real-time monitoring of the deposition process regulates heat accumulation within the substrate therefore enhancing the buildability of L-DED components.