Recently, breakthroughs of NLP research have improved a range of activities, most notably the Question Answering System for many languages. Since the last few years, question answering (QA) systems have grown at a breakneck pace. With the continuous development of the network, the question-and-answer method has become a way for people to get information quickly & precisely that the user will ask and with the increase in web sourcing, any information has become available to the people as the relevant data is stored in that source. LSTM has been introduced, a focus-based deep learning model for the Q&A method in this study. It matches one of the sentences in the question and answer and solves the problem of unexpected features. Using the attention mechanism in the system provides accurate answers by focusing on the specific questions of the candidate. Furthermore, we have proposed an adequate knowledge addition-based framework for the Q&A method. This memory contains a nested word or character level encoder that handles problems outside the words in the dataset or some rare words. We compare both Bangla and English-based question-answer for the dataset domain based on International GK, Bangladesh GK, and Science & Technology. A Sequence to Sequence LSTM based question-and-answer system with a total number of 10,000 data has been proposed through an attention mechanism with (99.91 and 99.48) % accuracy for Bangla and English data, respectively. Overall, LSTM works perfectly for both Bengali and English and is the best Q&A model.