Track 2 · Building with models
Retrieval-Augmented Generation (RAG)
RAG is how you connect an LLM to your own documents without training a new model. This course teaches the whole path: prepare the content, retrieve the right evidence, place it in context, and measure whether the answer is grounded.
01
02
03
04
05
06
07
What is retrieval-augmented generation?
The core RAG loop, why it exists, and the difference between model knowledge, retrieved evidence, and generated answers.
The RAG retrieval pipeline
How documents move from ingestion to retrieval: parsing, cleaning, metadata, indexing, search, context assembly, and answer generation.
Chunking documents for RAG
Why chunk size, overlap, boundaries, titles, and metadata decide whether retrieval finds a useful piece of evidence.
Embeddings and vector search for RAG
How semantic search works, what vector databases store, and why nearest neighbor search is useful but not magic.
Hybrid search and reranking
Why keyword search still matters, how hybrid retrieval works, and how rerankers improve the final evidence set.
RAG prompts, context, and citations
How to put retrieved chunks into the prompt, ask for grounded answers, and make citations point to real sources.
Evaluating RAG systems
The practical eval loop for RAG: retrieval recall, answer faithfulness, freshness, latency, cost, and failure mode tracking.