The problem addressed in this paper is the development of multiple models for natural language processing and their integration into the web interface. For this, it is necessary to create a dataset for training and testing of the models for translation tasks. For solving this problem, two types of models are trained. LSTM (long short-term memory) model is used for sentence prediction, in other words, from source language tries to predict translation of the sentence to the targeted language. BERT (Bidirectional Encoder Representations from Transformers) model is used for automatic guessing of the source language, that is, it is doing sentence classification tasks. For training and testing of each of the models, a special dataset is created, and it is converted into the format that models accept. All models are built on the Google Colab platform, which is a service that provides free access to the first-class hardware that can be used to make the training of the models faster. For testing the accuracy of the LSTM model used for translation task BLEU (Bilingual Evaluation Understudy) score is used which tells us how big a difference between automatic machine translation and one or more human-created references of the same source sentence is. The user interface is built using the Django web framework which is based on a Python programming language. The interface enables users to input source sentences, choose the language in which that sentence is written, and choose the language on which to translate that sentence. One of the options when choosing the source language is the option “auto”. In this case application, itself will try to guess the source language, and based on that it will choose the right LSTM model for translation of that sentence. The application allows users to translate text from all the given languages to Croatian and from Croatian to all given languages.