ID documents submitted for Maritime digital KYC processes can be skewed due to the environment in which the photograph is taken or due to user preferences and/or errors. The skewed image results in a low accuracy in downstream image processing tasks like optical character recognition (OCR). ID document deskewing has been typically approached using deep learning (MaskRCNN), regression, projection plans, Hough transforms, Fourier transforms and other computer vision techniques. The aim of this study is to build a robust document deskewing system based on keyword detection and coordinate geometry. The research is carried out by analyzing skewed Indian PAN cards available with IN-D. The database has 50 Indian PAN card images. These images are augmented to generate 150 images, with 50 images for each of the +90, -90 and 180 degree skew cases. Google Vision API is used as the OCR engine for finding the coordinates of the keyword in our study. The research employs Numpy, Pandas and OpenCV open-source libraries for Python. The accuracy of the reported model is 95.33 %. The accuracy of our present approach surpasses the accuracy of all the models available in literature.