BUILDING AN ENGLISH VOCABULARY LEARNING APPLICATION WITH LISTENING AND PRONUNCIATION PRACTICE

Abstract

Currently, searching and cross-referencing information in Vietnamese administrative documents is mainly done manually, which is time-consuming, prone to missing important content, and carries the risk of misapplying regulations, especially when handling multiple complex hierarchical documents simultaneously. The thesis develops a chatbot system for querying administrative documents based on the Retrieval-Augmented Generation (RAG) architecture. The system supports uploading multiple documents, automatically extracting text, performing logical chunking suitable for regulatory documents, generating semantic embeddings, indexing, and conducting efficient semantic search. Upon receiving natural language queries, the system retrieves relevant passages, refines result ranking, and generates responses entirely based on the original content, eliminating the hallucination phenomenon. The system runs locally with a user-friendly web interface, integrating features for user management, document management, chat history, and data backup. Testing results on real administrative documents demonstrate fast response times, high accuracy, low hardware requirements, and suitability for practical deployment.