This paper describes our submission to the WMT 2025 (Pakray et al, 2025) Shared Task on Low-Resource Machine Translation for Indic languages. This task is an extension of the efforts which was originally initiated in WMT 2023 (Pal et al., 2023), and further continued to WMT 2024 (Pakray et al, 2024), received significant participation from the global community. We address English ↔ {Assamese, Bodo, Manipuri} translation, leveraging Hindi and Bengali as high-resource bridge languages. Our approach employs Transformer-based Neural Machine Translation (NMT) models, initialized through multilingual pre-training on high-resource Indic languages, followed by fine-tuning on limited parallel data for the target low-resource languages. The pre-training stage provides a strong multilingual representation space, while fine-tuning enables adaptation to specific linguistic characteristics of the target languages. We also apply consistent preprocessing, including tokenization, true casing, and subword segmentation (Sennrich et al., 2016) with Byte-Pair Encoding (BPE), to handle the morphological complexity of Indic languages. Evaluation on the shared task test sets demonstrates that pre-training followed by fine-tuning yields notable improvements over models trained solely on the target language data.