How terry-xu-666/noderag works
Project type and purpose
• NodeRAG is a Python-based, full-stack RAG (Retrieval-Augmented Generation) system.
• It provides:
• A back-end library for building and querying a heterogeneous graph of “nodes.”
• A document ingestion/indexing pipeline.
• A retrieval & answer-generation pipeline (search + LLM integration).
• Visualization tooling (static HTML & WebUI).
File structure
Top-level modules:
• NodeRAG/build – document-to-graph ingestion pipeline (text splitting, embedding, graph construction, incremental updates).
• NodeRAG/search – graph retrieval logic.
• NodeRAG/LLM – LLM wrappers, routing, and state management.
• NodeRAG/storage – graph persistence and ID/graph mappings.
• NodeRAG/Vis – standalone HTML visualizations.
• NodeRAG/WebUI – a Python-based web interface (app.py).
• NodeRAG/config – YAML & Python config.
• NodeRAG/logging – standardized logging.
• NodeRAG/utils – helper functions (HNSW, PPR, prompts, YAML ops, token utils, etc.).
README for additional insights
• The system supports two primary flows:
a. Indexing (ingesting raw text → heterogeneous graph)
b. Answering (user query → retrieval → LLM generation)
• It offers both a WebUI and HTML-based visual exploration tool.
• Emphasizes heterogeneous node types, incremental graph updates, and explainable retrieval.
Elements
• Clear labels for each component (e.g., “TextSplitter”, “HNSWGraphBuilder”, “LLMRouter”)