In the quest for advancing computational tools capable of accurately calculating, estimating, or predicting partial atomic charges in organic molecules, this work introduces a pioneering Machine Learning-based tool designed to transcend the limitations of traditional methods like DFT, Mulliken, and semi-empirical approaches such as MOPAC and Gaussian. Recognizing the crucial role of partial atomic charges in molecular dynamics simulations for studying solvation, protein interactions, substrate interactions, and membrane permeability, we aim to introduce a tool that not only offers enhanced computational efficiency but also extends the predictive capabilities to molecules larger than those in the QM9 dataset, traditionally analyzed using Mulliken charges. Employing a novel neural network architecture adept at learning graph properties and, by extension, the characteristics of organic molecules, this study presents a "sliding window" technique. This method segments larger molecules into smaller, manageable substructures for charge prediction, significantly reducing computational demands and processing times. Our results highlight the model's predictive accuracy for unseen molecules from the QM9 database and its successful application to the resveratrol molecule, providing insights into the hydrogen-donating capabilities of CH groups in aromatic rings—a feature not predicted by existing tools like CGenFF or ATB but supported by literature. This breakthrough not only presents a novel alternative for determining partial atomic charges in computational chemistry but also underscores the potential of convolutional neural networks to discern molecular features based on stoichiometry and geometric configuration. Such advancements hint at the future possibility of designing molecules with desired charge sequences, promising a transformative impact on drug discovery.