This paper presents a novel approach to machine translation (MT) for the low resource language pair Sanskrit-Malayalam, leveraging deep learning techniques. The complexity of Sanskrit grammar and the relative paucity of computational resources for Malayalam pose unique challenges. By constructing a robust parallel corpus and employing state-of-the-art neural network architectures, the paper demonstrates significant improvements in translation quality compared to traditional rule-based and statistical methods. The promised work mainly focused on the translation of Sanskrit Shlokas to Malayalam using seq2seq model with attention mechanism, transformer-based and transfer learning approaches. A parallel corpus is created from ancient textbooks like Ashtanga Hrudaya, Bhagavad Gita, and Ramayana for training and testing.