Brain metastases are the most frequently diagnosed brain tumors and are associated with significant clinical and economic burdens. Accurate segmentation of metastases is essential for minimally invasive treatment but remains a time-consuming and labor-intensive process. This work presents a machine learning-based segmentation pipeline to compete in the Brain Tumor Segmentation (BraTS) Metastases 2025 Challenge, using the MONAI Auto3DSeg framework with three neural network architectures: SegResNet, DiNTS, and Swin-UNETR. The model was trained on 1,296 cases, each including four sequences (T1, T1-post contrast, T2, and FLAIR), with an NVIDIA ADA 6000 GPU. Training was done with two-fold cross-validation on 100 epochs with early stopping after 5 epochs. Once training was completed, the models each made predictions on 179 unlabeled cases, which were then post-processed by removing noise, enforcing a label hierarchy, and filling in holes. Then, the final predictions were submitted to the BraTS Synapse page for model evaluation. The final predictions achieved Dice Similarity Coefficients (DSC) of 0.87 for the resection cavity, 0.69 for the tumor core, 0.68 for the whole tumor, and 0.66 for the enhancing tumor. These results highlight the combination of multiple architectures and a robust post-processing pipeline to improve segmentation accuracy. Future work will focus on using more cross-validation folds, optimizing model architectures through neural architecture search, and expanding training data to improve the detection of smaller tumor regions.