The recognition of Hindi handwritten words is still a difficult issue because of the absence of big, high-quality datasets, a variety of handwriting styles, and restricted resources. On handwritten Hindi text, general-purpose OCR algorithms perform poorly, even with intensive multilingual training. This study uses a ResNet-50 convolutional neural network, trained on a publicly accessible Hindi handwriting dataset, to suggest a straightforward yet efficient supervised way. By contrasting our results with a popular commercial OCR system, we show that domain-specific, focused training greatly improves recognition performance in low-resource languages. In order to develop strong handwriting recognition systems, the results highlight importance of specialized datasets and low processing power