The community member is interested in using multi-agents for invoice data extraction, and wants to know if they can extend their existing standard rag pipeline to be "agentic". A comment explains that "agentic" means adding decisions that the language model makes, and provides examples such as using the language model to write queries and metadata filters for retrieval, or using an agent loop for multi-hop retrieval. However, there is no explicitly marked answer in the provided information.
Hello all, currently I want to investigate the use of multi-agents for my use case of invoice data extraction. I already have a standard rag pipeline set up. Just wondering if I can extend my application and make it agentic, and if yes, how can I do that?
I searched the internet but I can’t find much information about it. It’s all about training models and I don’t want those solutions.
"Agentic" just means adding decisions that the llm makes
What and where this happens is totally up to you. It could be something like using an llm to write queries and metadata filters for retrieval, and could be using an agent loop for multi-hop retrieval, etc.
