DeepCoder solves simplest competitive style programming problems from input-output examples using deep learning. It uses the principle of Learning Inductive Program Synthesis to induce programs that are consistent with given input-output examples. It uses a neural network to predict the probabilities of a program that generated the outputs from the inputs and use these predictions to augment the search technique, an enumerative search, to make the process faster. The results show an order of speedup over non-augmented approaches. This project can solve competitive programming style problems of the simplest level. However, since it is limited by the DSL, it cannot yet be applied to complex problems that require more complex approaches.