Bob DuCharme
4.9K posts
@bobdc
Senior Technical Writer at Graphwise (opinions mine, not theirs). Wrote O'Reilly's "Learning SPARQL." Follows you. Mastodon: https://t.co/dF1rZ2QqgT
Goodbye, vanilla RAG. Hello, Agentic RAG! 𝗩𝗮𝗻𝗶𝗹𝗹𝗮 𝗥𝗔𝗚 The common vanilla RAG implementation processed the user query through a retrieval and generation pipeline to generate a response grounded in external knowledge. Advanced vanilla RAG techniques include e.g., incorporation of rerankers to improve the retrieved results. 𝗧𝗵𝗲 𝗶𝘀𝘀𝘂𝗲? - Lack of flexibility: You retrieve additional information from a predefined knowledge source. - No data validation: Vanilla RAG systems are one-shot retrievers. There’s no validation of whether the retrieved data is even relevant to the question. 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗥𝗔𝗚 Agentic RAG is an agent-based approach to vanilla RAG. It involves the use of AI agents, especially in the retrieval component (as well as other components). The use of AI agents in the retrieval component enables tool use to generalize retrieval. 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗮𝗻𝗱 𝗞𝗲𝘆 𝗖𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁𝘀 • Retrieval Agent: Decides whether external knowledge is needed, where to retrieve data from, preprocesses the query for optimized retrieval, validates whether the retrieved data is relevant to the user query, and re-retrieves data if necessary. • Tool use: Data can be retrieved from multiple sources, including (vector) databases, web searches, API calls to email or chat messages, calculators, etc. In our recent blog @ecardenas300 and I discuss everything related to Agentic RAG. Read more on our blog: weaviate.io/blog/what-is-a…
⚡ Ontotext + Semantic Web Company = Graphwise ⚡ We are excited to announce the merger of Ontotext & @semwebcompany, forming a new entity called Graphwise. graphwise.ai Stay tuned for more details!
