{"id":51496,"date":"2024-12-23T11:39:58","date_gmt":"2024-12-23T11:39:58","guid":{"rendered":"https:\/\/www.bravent.net\/?p=51496"},"modified":"2024-12-30T09:09:23","modified_gmt":"2024-12-30T09:09:23","slug":"graphrag-introduction","status":"publish","type":"post","link":"https:\/\/www.bravent.net\/en\/news\/graphrag-introduction","title":{"rendered":"GraphRAG Introduction"},"content":{"rendered":"

GraphRAG<\/a> is an innovative technique that combines the power of LLM models with knowledge graphs to structure and analyze unstructured\/organized information. Aimed at processing and understanding large volumes of documents, reports and data into actionable ideas or insights, it is ideal for companies and researchers working with large volumes of data and documents or complex information.
\nLet’s start by looking at what RAG Retrieval-Augmented Generation is and see the fundamentals and differences with GraphRAG.<\/p>\n<\/div>[vc_empty_space]

RAG Retieval-Augmented Generation<\/h2>\n

LLMs are models that use generative AI to create content, images, code from existing data, data on which they are trained but have a limitation, they are finite, they are generalist and not always the answers they give us are those we expect or need to cover our specific use case, that is, we need to give context or train it on our data so that the answers they give us are the ones we need to obtain or give to our customers.<\/p>\n

To solve this limitation, the pattern called RAG Retrieval-Augmented Generation arose, it is the technique that allows us to give context to our model and that its response is based on our data or information.<\/p>\n

This technique consists of storing the client’s information in vector databases, so that when a user makes a query to the LLM, the first thing we do is to look for the information related to the query in the database and send it to the AI so that the answer it gives us is about that information or data.<\/p>\n

Traditional RAG systems have found applications in a variety of domains including:<\/p>\n