Accurate segmentation of glioma subregions in brain MRI remains a critical challenge due to high inter-subject variability and complex tumor morphology. In our submission to the BraTS 2025 challenge, we develop a pipeline based on nnU-Net, enhanced with domain adaptation, pseudo-labeling, and custom post-processing. Our method focuses on addressing distributional shifts across modalities and scanners, while improving lesion consistency through connected component filtering. The approach demonstrates moderate performance on the validation set, with strong lesion-wise Dice scores. Ongoing improvements include the integration of boundary-aware loss functions and surface-based loss functions to mitigate residual segmentation artifacts, particularly along tumor borders. To make the process more accessible and reproducible, we focus on optimizing the pipeline for consumer-level hardware. The segmentation is performed in 2D configuration, which reduces computational overhead, making it more feasible for users with limited resources. However, while the current approach is lightweight, ongoing improvements, including the integration of boundary-aware loss functions and surface-based loss functions, are being made to address residual segmentation artifacts along tumor borders. These enhancements involve more computationally intensive tasks like Distance transform maps (DTM) calculation, which can require significant computation time, but are critical for further refinement of the segmentation accuracy and convergence.