The project is concerned with building an intelligent chatbot on the field of cryptocurrency or other hot fields via deep learning. BERT is one such pre-trained model developed by Google which can be fine-tuned to new data and used to create NLP systems like question answering, text generation, text classification, text summarization, and sentiment analysis. As BERT is pretrained on huge amounts of data, the process of language modeling is easier. The main benefit for using a pre-trained model of BERT is substantial accuracy improvement compared to training on these datasets from scratch. The central question that this project asks, then, is: how to improve the results of chatbot? A few steps we have been doing: collected data via scraping quora and related websites and pre-processed the data, then used BERT to fine-tune data. We also implemented chatbot based on BERT. In addition, we tuned up different hyperparameters to obtain the optimal results. Next, we explore whether a new hybrid deep network model can predict the next statement and how it will perform. We explore new algorithms or modify the existing algorithms to improve the chatbot performance. Future work will explore one-shot learning algorithm to perform sentiment analysis on our chatbot.