It is no secret that AI is an upcoming titan. Even though people are stunned to hear that AI has been here for around a century, due to the advancement in computational methods and resources, today AI peaks like never before. As a tiny glimpse into the field of Digit Recognition, this project aims to understand the underlying cogs and wheels on which the neural networks spin. This paper tries to elucidate a project which solves the Sudoku puzzle drawn and written by hand. The paraphernalia for that project includes programming language: Python3; libraries: OpenCV, Numpy, Keras; datasets: MNIST handwritten digit database. Digit recognition is a classical problem which will introduce neurons, neural networks, connections hidden layers, weights, biases, activation functions like sigmoid, back-propagation and other related topics as well. Algorithm(s) in the project employed to solve Sudoku is also explored in this paper.