Named Entity Recognition (NER) in Hindi documents poses unique challenges due to the language’s rich morphology and syntax. This research investigates the application of the pre-trained XLM-RoBERTa model for NER tasks in Hindi. Leveraging transfer learning, the study fine-tunes XLM-RoBERTa on the HiNER dataset, which comprises annotated sentences from diverse sources. The model demonstrates exceptional performance across all metrics. It achieved a low evaluation loss of 0.0854, a high recall, precision and an F1 score of 89.9%, indicating a strong ability to accurately identify positive cases while minimizing false positives. Additionally, the model boasts an impressive overall accuracy of 97.3%. These metrics show how well the model recognizes and classifies named things in Hindi. The paper provides important insights into the application of advanced transformer-based models for natural language processing tasks and emphasizes their promise in enhancing NER for underrepresented Hindi language.