Question-answering (QA) systems have gained immense popularity in the recent past due to their extensive usage in designing Chat- Bots and Kiosks. Developing interactive QA systems has always been a complex task in NLP. Time to look at QA systems now! These are available in different standards. However, this work focuses on designing an extractive QA system, which finds an answer-span from a con- text paragraph. Researchers have proposed many domain-specific QA systems to meet their needs. Yet high-quality QA systems are scarce for answering narrative text content from a collection of the document corpus. On the contrary, the BERT model has proved to be a powerful pre-trained language model and can be ne-tuned for developing efficient domain QA systems. In this paper, we develop an open-domain QA sys- tem based on BERT, termed Narrative QA-BERT (NQA-BERT). Particularly, the proposed NQA-BERT is pre-trained on the NarrativeQA dataset and ne-tuned on document corpus demonstrating the Indian epic stories such as Ramayana and Mahabharata. The effectiveness of the intended NQA-BERT model is measured with different performance metrics. The experimental results demonstrate that the ne-tuned BERT achieves high accuracy in extracting the answer-spans.