This paper introduces the Bi-Directional Attention Flow (BiDAF) network, a multi-stage hierarchical process designed to improve machine comprehension (MC) tasks. BiDAF represents the context at different granularities and uses a bi-directional attention flow mechanism to achieve a query-aware context representation without early summarization. The model includes character-level, word-level, and contextual embeddings, and employs a memory-less attention mechanism that allows attention vectors to flow through subsequent modeling layers, reducing information loss. BiDAF also uses bidirectional attention in both directions (query-to-context and context-to-query) to provide complementary information. Experimental results show that BiDAF outperforms previous approaches on the Stanford Question Answering Dataset (SQuAD) and the CNN/DailyMail cloze test, achieving state-of-the-art results. The paper includes ablation studies, visualizations, and error analysis to demonstrate the effectiveness of each component in the model.This paper introduces the Bi-Directional Attention Flow (BiDAF) network, a multi-stage hierarchical process designed to improve machine comprehension (MC) tasks. BiDAF represents the context at different granularities and uses a bi-directional attention flow mechanism to achieve a query-aware context representation without early summarization. The model includes character-level, word-level, and contextual embeddings, and employs a memory-less attention mechanism that allows attention vectors to flow through subsequent modeling layers, reducing information loss. BiDAF also uses bidirectional attention in both directions (query-to-context and context-to-query) to provide complementary information. Experimental results show that BiDAF outperforms previous approaches on the Stanford Question Answering Dataset (SQuAD) and the CNN/DailyMail cloze test, achieving state-of-the-art results. The paper includes ablation studies, visualizations, and error analysis to demonstrate the effectiveness of each component in the model.