GraphRAG Studio
A full-stack GraphRAG application with a Django backend and React frontend that builds a knowledge graph from ingested documents, detects thematic communities, and answers questions using hybrid vector and graph retrieval with LLM-powered reasoning.
GraphRAG Studio extracts entities and their relationships from ingested text documents to build a knowledge graph, then groups related entities into communities and uses an LLM to generate summaries for each group, giving both a fact-level and a theme-level view of the data that plain vector-similarity RAG tends to lose on broad, multi-hop questions.
Answering a question runs a multi-step pipeline: entity extraction expands the question across the graph to include strongly related neighboring entities so relevant context is not missed; a hybrid retrieval step combines dense vector embeddings with BM25 keyword search for local matches and compares the question against precomputed community summaries for global, thematic matches; results from both are merged, deduplicated and reranked with a cross-encoder through reciprocal rank fusion; and the LLM synthesizes a final answer from the highest-ranked context.
It ships as a full-stack application with a Django backend and a React frontend.