Agentic RAG

#RAG

Agentic RAG

Recently, I have seen lots of posts and videos about agentic RAG.
Like,

I’m excited to kick off the first of our short courses focused on agents, starting with Building Agentic RAG with LlamaIndex, taught by @jerryjliu0, CEO of @llama_index.

This covers an important shift in RAG (retrieval augmented generation), in which rather than having the… pic.twitter.com/KwRvDEhENS

— Andrew Ng (@AndrewYNg) May 8, 2024
, and

If you want to learn about building agentic RAG systems this weekend, then we recommend looking into this comprehensive blog/tutorial series by @Prince_krampah - starting with basic routing and function calling to multi-step reasoning over complex documents 🧠📑

Best of all, it… pic.twitter.com/YTx7MDu2X7

— LlamaIndex 🦙 (@llama_index) June 8, 2024

Pasted image 20240622043159.png

Basically, it is a combination of agent&RAG. Classic/Vanilla RAG works with vector store or knowledge graph in a cascade manner/one-shot query, while agentic RAG introduces agent reasoning loop into the process.

Specifically, agentic RAG is a framework that uses intelligent agents to improve information retrieval and generation. It involves a meta-agent coordinating document agents, each capable of understanding and summarizing their documents, to provide comprehensive answers to complex queries.

  1. Document Agents: Each document is associated with a document agent capable of answering questions and summarizing content within its own document.
  2. Meta-Agent: A top-level agent, known as the meta-agent, manages all the document agents. It orchestrates the process by deciding which document agent to use for a given query.
  3. Query Processing: The user’s query is processed by the meta-agent to determine the most relevant document agents.
  4. Information Retrieval: The selected document agents retrieve information from their respective documents.
  5. Synthesis: The meta-agent synthesizes the retrieved information to generate a comprehensive response.

Pasted image 20240622034047.png

In this case, there are two documents/papers.
Pasted image 20240622034851.png

A simple example of creating vector query and summarizing tools for a doc. 👇

async def create_doc_tools(  
document_fp: str,  
doc_name: str,  
verbose: bool = True,  
) -> Tuple[QueryEngineTool, QueryEngineTool]:  
# load lora_paper.pdf documents  
documents = SimpleDirectoryReader(input_files=[document_fp]).load_data()  
  
# chunk_size of 1024 is a good default value  
splitter = SentenceSplitter(chunk_size=1024)  
# Create nodes from documents  
nodes = splitter.get_nodes_from_documents(documents)  
  
# LLM model  
Settings.llm = OpenAI(model="gpt-3.5-turbo")  
# embedding model  
Settings.embed_model = OpenAIEmbedding(model="text-embedding-ada-002")  
  
# summary index  
summary_index = SummaryIndex(nodes)  
# vector store index  
vector_index = VectorStoreIndex(nodes)  
  
# summary query engine  
summary_query_engine = summary_index.as_query_engine(  
response_mode="tree_summarize",  
use_async=True,  
)  
  
# vector query engine  
vector_query_engine = vector_index.as_query_engine()  
  
summary_tool = QueryEngineTool.from_defaults(  
name=f"{doc_name}_summary_query_engine_tool",  
query_engine=summary_query_engine,  
description=(  
f"Useful for summarization questions related to the {doc_name}."  
),  
)  
  
vector_tool = QueryEngineTool.from_defaults(  
name=f"{doc_name}_vector_query_engine_tool",  
query_engine=vector_query_engine,  
description=(  
f"Useful for retrieving specific context from the the {doc_name}."  
),  
)  
  
return vector_tool, summary_tool