A basic RAG pipeline works well on simple questions, but it can struggle when questions are vague, when the retrieved passages are off-topic, or when the answer is spread across many documents. Advanced RAG adds steps before, during and after retrieval to handle these cases.
You will start with an overview of the main techniques, then look at query rewriting, which reshapes the user's question so the search finds better passages. Next comes self-correcting RAG, where the system checks its own retrieved documents and answers, and searches again if they are not good enough. The last video covers Graph RAG, which builds a knowledge graph from your documents to answer questions that need the bigger picture.
These techniques add cost and complexity, so treat them as tools to reach for when a basic pipeline falls short, not as defaults.
An overview of advanced RAG concepts that go beyond a basic retrieve-and-generate pipeline, from the Complete Generative AI Course by AmanAI Lab.
Video 19minBuilds an adaptive, self-reflective RAG app with LangGraph, Groq, Chroma and Tavily web search, where the system checks its own results and corrects itself.
Video 13minHow the Microsoft GraphRAG system builds a knowledge graph with summarized communities from your documents, how to run it locally, and what it costs compared with traditional RAG.
Video 16min