This project implements a medical intelligent Q&A system using knowledge graph-based RAG with large language models. It builds a Neo4j knowledge graph containing 44,000 entities and 310,000 relations from a medical dataset. The system uses BERT plus RNN for named entity recognition achieving 97.40% F1 score, and employs local LLMs through ollama for 16-class intent recognition and streaming answer generation. A Streamlit interface provides the interactive frontend for the full pipeline from entity extraction to knowledge graph Cypher queries to final answer synthesis.