Adding Agentic Layers to RAG

In this talk, Jerry Liu, Co-Founder of LlamaIndex, dives into the world of Retrieval Augmented Generation (RAG) and discusses how to incorporate agents into the RAG framework. He introduces LlamaIndex, a data framework for building LLM applications, and explains the limitations of naive RAG prototypes. Jerry explores the challenges with naive RAG and presents solutions for handling complex questions, including summarization, comparison, structured analytics, and multipart inquiries. He further delves into the concept of agents, their role in utilizing LLMs for automated reasoning and tool selection, and the different layers at which they can be added to the RAG pipeline. From basic routing and query planning to tool use and agentic loops, Jerry showcases a range of agentic reasoning methods. He also touches on the exciting possibilities of long-term planning agents that optimize system-level components. Throughout the talk, he emphasizes the importance of observability, control, and customizability in building effective agents. If you're interested in understanding how to enhance your RAG applications with agentic layers and explore various agent paradigms, this talk provides valuable insights and practical tips. Check out LlamaIndex's documentation for more information. #AI #RAG #Agents

[@user-me7xe2ux5m]

Awesome presentation. Clear, well-structured, and easy-to-follow. Love the term "Dynamic QA System" rather than bucketing all internal knowledge query use-cases into RAG. Adding agents at different locations of a vanilla RAG workflow seems to be a powerful system architecture for solving a large set of QA use cases. Lots of food for thought!

[@nexuslux]

Nice presentation. Just didn’t see enough from llama index… but very well spoken and interesting.

[@johnwallis1626]

good presentation, my issue is using langchain/llama in production, they just add another unnecessary and buggy layer, maybe things will change moving forward, also adding layers of agents can ramp up costs quite significantly, which is where need for good open source llms comes in.

[@heythere6390]

But why is llamaindex so shitty then?