One of the main NLU tasks is to understand the intents (sequence classification) and slots (entities within the sequence). This repo help classify both together using Joint Model (multitask model). BERT_SMALL is used which can be changed to any other BERT variant.

Finding look alike sentences by leveraging the concept of semantic similarities pre-learned by transformer models while pre-training task. I've used cosine similarity as an angular distance matrix applied over sent2vec.