Although Large Language Models (LLMs) have made significant progress in Natural Language Processing the lack of high-quality training data frequently limits their ability to perform well in regional languages. To improve LLM competency this study methodically translates an English dataset into the low-resource language of Bhojpuri. On this new dataset we apply a structured translation methodology and then refine an LLM that has already been trained. The model’s capacity to produce contextually relevant and culturally appropriate responses in Bhojpuri has significantly improved according to a comparison of its performance before and after fine-tuning. Our findings show that this translation-centric approach provides a practical and affordable way to enhance the usefulness and inclusivity of LLMs increasing the effectiveness and accessibility of these potent AI tools for underrepresented linguistic groups globally. For linguistic groups that are marginalized globally.