Methodologies to leverage LLMs for Document Q&A

There are a number of different approaches that leverage LLMs for Document Q&A. Below are four approaches, each with distinct advantages and trade-offs.

• Simple LLM Workflow (this Blueprint): Simple LLM workflow works by extracting and splitting a document into structured sections and saving them separately. When a question is asked, the system identifies the most relevant section using an LLM and then generates an answer based on the retrieved content.
• Standard RAG: Standard RAG combines the capabilities of LLMs with an external knowledge base to enhance response accuracy. It operates in two phases: first, it retrieves relevant documents based on user queries using vector embeddings; then, it augments these documents with the original query and generates a response.
• Agentic RAG: Agentic RAG extends standard RAG by incorporating intelligent AI agents that actively reason and optimize the information retrieval process. These agents can break down complex queries, evaluate multiple sources, and validate information before presenting it.
• ‍Long-context model: Long-context models are AI systems capable of processing extensive amounts of text in a single input, often handling contexts of hundreds of thousands or even millions of tokens.