Description
Q&A over long videos or many documents with plain vector search returns scattered pieces and loses context. Context-Aware RAG is NVIDIA's RAG library that ingests content into a knowledge graph before retrieval.
It supports GraphRAG and agentic retrieval, used for long-video understanding and as a core of the video search and summarization blueprint.
Graph ingestion:Context preserved.
GraphRAG:Graph retrieval.
Agentic:Multi-step reasoning.
Long video:Understanding and summaries.
It supports GraphRAG and agentic retrieval, used for long-video understanding and as a core of the video search and summarization blueprint.
Features
Graph ingestion:Context preserved.
GraphRAG:Graph retrieval.
Agentic:Multi-step reasoning.
Long video:Understanding and summaries.
