Recurrent Neural Networks are surprisingly effective. They are capable of solving many problems which usually involve some sort of sequence. This might be the sequence of characters, sequence of words, the notes, the elements of energy time series describing the electrical energy consumption and many other examples. The common denominator of all of them is the sequence nature of the quantity. The chemical compound, being in principle the three-dimensional object, can be successfully and uniquely encoded as a sequence of characters. Relatively simple syntax of SMILES code allows on fast and comprehensive translation of complicated structure of the molecule into convenient short-hand notation. Within this presentation a set of biologically active species have been turned into the SMILES character representation. Such a representation have been passed to particular form of the Recurrent Neural Network. Carefully converged neural Network has been later on "stimulated" to produce other molecules. It turns out that these molecules are, to large extent, syntactically correct. It is also postulated that together with the knowledge related to the proper syntax of the molecule, also the semantics - i.e. the biological activeness - has also been learned. This would mean that new molecules are expected to pose similar biological properties. There are strong indications that it is indeed the case. Entire workflow of this application has been developed within the Python Ecosystem. The audience will have an opportunity to get familiar with theoretical background of Recurrent Neural Network and the cheminformatice tools that have been used to create relevant data sets. Within the final part of the presentation the pretrained model will be used to generate new molecules.