Skip to content

Advanced RAG

A production-quality course covering Retrieval-Augmented Generation from the ground up — from the basic loader → splitter → retriever → LLM pipeline to advanced patterns like RAG Fusion, HyDE, Agentic RAG, Graph RAG, and RAGAS evaluation.

Based on the Advanced RAG course by CampusX. Reference code: campusx-official/Advanced_Rag_Codes.

Why this course exists

LLMs without retrieval have four hard limits:

  1. Knowledge cutoff — models are frozen in time, can't access recent information.
  2. Hallucinations — they generate plausible-sounding but false content.
  3. No source attribution — you can't verify or trace where information came from.
  4. No private data access — they can't read your company's documents.

RAG fixes all four by grounding the LLM in external, retrievable knowledge at query time.

The standard RAG pipeline

flowchart LR
    A[User Query] --> B[Embedder]
    B --> C[Vector Store]
    C --> D[Retriever]
    D --> E[Top-k Docs]
    E --> F[Prompt Template]
    A --> F
    F --> G[LLM]
    G --> H[Answer]

Knowledge ingestion happens once, offline:

flowchart LR
    A[Raw Docs] --> B[Loader]
    B --> C[Splitter]
    C --> D[Chunks]
    D --> E[Embedder]
    E --> F[Vector Store]

This course covers each box, then builds the advanced patterns on top.

Syllabus

# Chapter What you learn
1 Introduction to RAG The four LLM problems, the RAG fix, full pipeline overview
2 Document Loaders PDFs, HTML, websites, databases — getting data into LangChain
3 Text Splitters Chunking strategies, recursive/markdown/code splitters, overlap
4 Text Embeddings Vectors, semantic similarity, OpenAI/HuggingFace embeddings
5 Vector Stores FAISS, Chroma, pgvector, persistence, indexing
6 Retrievers Similarity, MMR, threshold retrievers — the bread and butter
7 Advanced Retrievers Multi-query, contextual compression, parent-doc, self-query, ensemble
8 RAG Fusion Sub-query generation, Reciprocal Rank Fusion, ensemble retrieval
9 HyDE RAG Hypothetical document embeddings to bridge query/doc style gap
10 Agentic RAG LLM as router/orchestrator over multiple knowledge sources + tools
11 Graph RAG Knowledge graphs in Neo4j, entity extraction, hybrid retrieval
12 RAGAS Evaluation Faithfulness, answer relevance, context precision/recall metrics

Prerequisites

What you'll need

Tool What for
Python 3.10+ Runtime
langchain + langchain-openai The framework
OpenAI API key (or Anthropic / local model) LLM + embeddings
FAISS or Chroma Local vector store for dev
Neo4j Aura (free tier) Only for the Graph RAG chapter
ragas Only for the evaluation chapter

Most chapters work with a pip install and a free API key. Graph RAG needs Neo4j; RAGAS needs an LLM for the judge.

→ Start: Introduction to RAG