RAGViz in AI answers

RAGViz is an open-source tool that visualizes retrieval-augmented generation by showing both document- and token-level attention on the retrieved context fed to an LLM. It allows adding or removing documents to compare how different retrieved sources influence generated tokens, enabling diagnosis of the effectiveness and influence of retrieved documents. The project supports configurable datasets and LLM backbones via HuggingFace, uses AnchorDR embeddings and DiskANN indexes for retrieval, and provides a frontend UI and demo.