Abstract: The process of extracting textual data from unstructured data i.e., images is termed as Optical Character Recognition. This technology is widely employed in many different industries, including banking, healthcare, education, and document processing. The accuracy of the process is greatly affected by the kind of data that is available, as OCR models are sensitive to features of the document like the font, character spacing, quality of the scanned and handwritten documents. To improve the accuracy, OCR and Natural Language Processing (NLP) can be combined. When NLP is used with OCR, it can help to fix mistakes in the text that is recognised and raise the overall accuracy of the outcome. Utilising a language model to forecast the most likely words and phrases in the text is one popular method. In this document, we propose an OCR-NLP approach using language model T5 trained on the JFLEG dataset, improving the accuracy of text recognized by OCRs trained on the IAM and A-Z dataset. In this instance, a loss of 0.4571 indicates that the predictive performance of the correction model is optimal.