Many diverse day-to-day applications, such as e-commerce, marketing, supply chain, customer relationship management, and so on, have a significant impact on the Question-Answering (QA) System., It also creates a vivid emphasis among teaching learning systems, where the need for replies to inquiries is intrinsically required for e-learning application stakeholders. The authors of this paper present a deep learning approach to extract lecture notes as Texts and predicts the best acceptable or effective answer to a questionnaire.We used a pre-built BERT base model that has been fine-tuned with the Stanford Question Answering Dataset (SQuAD). The suggested transfer learning approach, we claim, aids generalizations in QA without the need of vocabulary. This unique framework focuses on developing a context-based question learning system that is customizable and can be integrated into any e-learning platform to benefit both teachers and students. It also employs a common deep learning approach known as transfer learning to enhance its ability to associate multiple practical provinces.Using a custom dataset created for the course "Object Oriented Programming," we fine-tuned the BERT transformer model. The experiment was carried out, and the results were visualised through comparisons with several models. In the testing phase, the MAP (Mean Average Precision) and MRR (Mean Reciprocal Rank) performance matrices were used to compare against the baseline of 82 percent MAP and 80.1 percent exact match score.