🕸 Turn text into a "graph of concepts" and use it as a RAG retriever instead of a vector DB. An OSS you can run fully locally. Title: rahulnyk/knowledge_graph URL: 🔍 Overview A project that builds a knowledge graph from unstructured text by extracting concepts (not just named entities) and their relationships, for Graph-Augmented Generation (GRAG) and knowledge-based QA. 🧩 Problem Solved Traditional text analysis struggles to reveal how concepts interconnect and where hidden links lie. Representing knowledge as a queryable graph that preserves meaning enables deeper document understanding. 🛠 Methodology & Approach Built in six steps (clean → extract concepts → extract relations → schema → populate nodes/edges → visualize). Edges carry two weights: W1 for LLM-extracted explicit relations, W2 for same-chunk co-occurrence. Node degree and community set node size and color. 💻 Tech Stack ・LLM: Mistral 7B OpenOrca (no GPT API) ・Serving: Ollama for a fully local pipeline ・Graph: NetworkX, Visualization: Pyvis, Data: Pandas 🎯 Use Cases Graph RAG using the graph as a retriever instead of a vector DB, hidden-link discovery, centrality analysis, and community detection. #KnowledgeGraph# #GraphRAG#