Optical Character Recognition using Deep Learning

I have created a character level and a word level neural network to recognize hand-written text. I used 2 datasets for this purpose, EMNIST Dataset for character level recognition and IAM Handwriting Dataset for word level recognition. The character level model uses a CNN architecture and achieved an accuracy of 88% . The word level model uses a CNN architecture which feeds into a LSTM layer and this model achieved an accuracy of 79%.
Github