The Southern Wide-field Gamma-ray Observatory (SWGO) will be a new ground-based telescope planned for construction in the Southern Hemisphere. It is designed to map the Galactic Center, study gamma-ray and cosmic-ray physics, and explore physics beyond the Standard Model. The reconstruction of the primary energy and the shower core position of Extensive Air Showers (EAS) is a fundamental step for event characterization and subsequent analyses. This thesis investigates the use of Multilayer Perceptron (MLP) models, implemented in Lux Julia, to reconstruct these parameters within the context of SWGO. A simulated dataset of photon induced EAS, provided by the SWGO Collaboration, was used. The events were generated with CORSIKA, AERIE, and the SWGO-RECO reconstruction chain. The dataset covers pri mary energies between 101.5 and 106 GeV, following a power law E−2, and includes secondary particles detected by the D8 reference layout, composed of 3763 Water Cherenkov Detectors distributed across three concentric radial density regions (70/4/1.7)%. After applying standard quality cuts, a dataset of approximately 2.3M events was obtained. Preprocessing focused on hit-time and charge information from each PMT; feature engineering was then applied, and the outputs were finally normalized. Two independent MLP models were implemented for the two reconstruction tasks, sharing optimization strategies such as AdamW, a One-Cycle scheme with a cosine learning rate, inverse-cyclical momentum, and dropout regularization. The en ergy reconstruction reaches R2 = 0.933%, slightly outperforming the likelihood-based method (R2 = 0.916%). The improvements are notable at low energies (101.5–103.0 GeV). In the inter mediate range, only smaller improvements are observed, while at high energies no improvement is obtained with respect to the likelihood-based method. For core reconstruction, the MLP reaches a median positioning error of 7.4 m, slightly worse than the baseline value of 5.6 m. Overall, the MLP improves energy reconstruction, but does not yet achieve competitive performance for core prediction. Although further optimization and more refined event-selection cuts could improve the results, the current energy model provides a solid baseline for future developments.