KG_RAG

Why we included this project

Teams building retrieval-augmented systems around biomedical questions will find KG-RAG a useful reference. It pairs SPOKE, a knowledge graph with tens of millions of nodes and edges covering genes, drugs, diseases and how they connect, with a general-purpose LLM such as GPT or Llama, extracting just the prompt-aware context needed to answer a query. The notebooks and the BiomixQA benchmark show, side by side, how the same drug prompt changes when the graph context is added. Biomedical data ships with the repo and a setup script builds the vector index, so a working pipeline comes together quickly. The caveat: this release is tuned for disease and drug questions, so treat it as a solid template for life-sciences RAG rather than a general-purpose tool.