This work is part of the Large Scale Semantic Indexing and Question Answering project. Particularly, user poses a question and the system returns the top 10 related biomedical documents, top 10 passages that probably contain the answer and finally the most relevant answer. The system recognizes the type of the question (factoid or yes/no). Factoid QA component is a learning model on top of the Bert (language representation) while the Yes/No QA component is a learning model that incorporates sentiment information and ELMo embeddings.