The progress of neural machine translation has significantly enhanced the overall quality of the field. However, notable challenges persist, particularly concerning the fairness of output translations. Neural models are trained on extensive text corpora containing inherent biases and stereotypes, leading them to absorb these social biases. Recent methodologies have shown success in mitigating gender bias in various natural language processing tools, prompting our exploration of word embedding techniques to address gender bias in Neural Machine Translation systems. Our approach leverages word embeddings in neural machine translation to rectify gender biases. We developed our method by training on Hindi-English translation using the IIT-Bombay Hindi-English parallel corpus, consisting of approximately 1.5 million sentence pairs, and evaluated it on WinoMT[24] and WinoBias [35] datasets. We propose a novel debiasing approach utilizing pre-trained debiased word embeddings [30] specifically designed to mitigate biases. Our transformer-based NMT model underwent evaluation using metrics such as F1 score and other statistical measures. In summary, recent efforts have introduced training methods and inference techniques to mitigate harmful gender bias in machine translation. Our research contributes a new debiasing method and provides an analysis of multilingual Hindi-English translation.