Retrieval Augmented Generation (RAG) Course

Lesson 1 • Key Components and Terminology

Defines essential vocabulary including corpus, chunk, embedding, and vector store. Precise terminology prevents confusion in later technical chapters.

Lesson 2 • What RAG Solves

Identifies the knowledge limitations of static LLMs and how retrieval addresses them. Anchors the entire course by establishing the problem RAG is designed to solve.

Lesson 3 • Core RAG Architecture Overview

Maps the three-stage pipeline: indexing, retrieval, and generation. Gives students a mental model they will refine throughout the course.

Lesson 1 • Metadata Enrichment and Tagging

Explains attaching source, date, section, and custom tags to each chunk. Rich metadata enables filtered retrieval and improves answer attribution.

Lesson 2 • Chunking Strategies

Compares fixed-size, sentence-based, recursive, and semantic chunking methods. Chunk quality is the single largest controllable factor in retrieval performance.

Lesson 3 • Evaluating Chunk Quality

Introduces metrics and manual inspection methods to assess whether chunks are coherent and complete. Connects chunk quality directly to retrieval recall and precision.

Lesson 4 • Text Cleaning and Normalization

Teaches removal of noise such as headers, footers, and boilerplate before chunking. Clean text directly improves embedding quality and retrieval accuracy.

Lesson 1 • Selecting an Embedding Model

Guides model selection using benchmark scores, domain fit, latency, and cost. Correct model choice prevents costly re-indexing after deployment.

Lesson 2 • Similarity Search and Distance Metrics

Explains cosine similarity, dot product, and Euclidean distance and when to use each. Metric choice affects ranking quality and must match the embedding model's training objective.

Lesson 3 • Indexing and Upsert Workflows

Covers batch indexing, incremental upserts, and deletion to keep the vector store current. Stale indexes degrade retrieval quality in production systems.

Lesson 4 • Embedding Model Fundamentals

Explains how transformer-based encoders map text to fixed-dimensional vectors. Provides the theoretical basis for all retrieval techniques covered later.

Lesson 1 • Query Transformation Techniques

Covers query rewriting, expansion, HyDE, and multi-query generation to improve recall. Transformed queries surface relevant chunks that the original query would miss.

Lesson 2 • Re-ranking Retrieved Results

Applies cross-encoder re-rankers and LLM-based scoring to reorder initial candidates. Re-ranking places the most relevant chunks at the top of the context window.

Lesson 3 • Hybrid Retrieval and Fusion

Combines dense and sparse scores using reciprocal rank fusion and weighted blending. Hybrid retrieval consistently outperforms either method alone across diverse query types.

Lesson 4 • Sparse Retrieval with BM25

Introduces term-frequency-based BM25 as a complement to dense retrieval. Sparse methods excel on exact-match and rare-term queries where dense models underperform.

Lesson 1 • Context Injection Strategies

Compares stuffing, map-reduce, and refine patterns for inserting multiple retrieved chunks. Strategy choice affects coherence and token budget consumption.

Lesson 2 • Instruction Design for Faithfulness

Teaches instructions that constrain the model to answer only from provided context. Faithfulness constraints are the primary defense against hallucination in RAG.

Lesson 3 • Anatomy of a RAG Prompt

Breaks down system instructions, retrieved context, and user query into distinct prompt zones. Understanding prompt structure is prerequisite to systematic prompt optimization.

Lesson 4 • Token Budget Management

Addresses context window limits by prioritizing, truncating, and compressing retrieved content. Efficient token use ensures the most relevant information reaches the model.

Lesson 1 • Debugging and Root Cause Analysis

Provides a systematic process for tracing poor answers back to retrieval, chunking, or prompt failures. Root cause analysis prevents misattributing generation errors to the wrong pipeline stage.

Lesson 2 • Evaluation Dimensions in RAG

Defines the four axes of RAG quality: context relevance, faithfulness, answer relevance, and completeness. A multi-dimensional view prevents optimizing one metric at the expense of others.

Lesson 3 • End-to-End RAG Evaluation Frameworks

Introduces RAGAS and similar automated frameworks that score full RAG pipelines. Automated frameworks enable continuous evaluation as the system evolves.

Lesson 4 • Retrieval Evaluation Metrics

Covers precision, recall, MRR, and NDCG for measuring retrieval quality against labeled datasets. Retrieval metrics isolate indexing and search problems from generation failures.

Lesson 1 • Iterative and Multi-Step Retrieval

Chains multiple retrieval calls where each step refines or expands the query based on prior results. Multi-step retrieval handles complex questions requiring evidence from multiple sources.

Lesson 2 • Graph-Enhanced Retrieval

Augments vector retrieval with knowledge graph traversal to capture entity relationships. Graph context improves answers on questions requiring multi-hop reasoning.

Lesson 3 • Agentic RAG with Tool Use

Integrates retrieval as one tool among many in an LLM agent that plans and executes multi-action workflows. Agentic RAG enables dynamic decision-making beyond static retrieval pipelines.

Lesson 4 • Multimodal RAG

Extends RAG to retrieve and reason over images, tables, and mixed-media documents. Multimodal retrieval unlocks use cases where critical information is non-textual.

Lesson 1 • Observability and Monitoring

Instruments the pipeline with tracing, latency dashboards, and retrieval quality monitors. Observability enables rapid detection and resolution of production degradation.

Lesson 2 • Continuous Improvement in Production

Establishes feedback loops using user signals, shadow testing, and A/B experiments to improve the live system. Continuous improvement prevents quality decay as data and usage patterns evolve.

Lesson 3 • Security and Access Control

Implements document-level permissions, query sanitization, and output filtering in the RAG pipeline. Security controls prevent unauthorized data exposure through retrieval.

Lesson 4 • Scalability and Performance Tuning

Addresses horizontal scaling of retrieval, caching strategies, and batching to meet throughput targets. Performance tuning ensures acceptable latency under production query volumes.