Retrieval systems are at the core of modern Gen AI applications. But most explanations focus on tools, not understanding. This course teaches you how retrieval actually works, for RAG for LLMs, and how to design and evaluate it effectively.

You’ll learn how information is represented through embeddings, how similarity is computed, and why similarity alone is not enough to guarantee relevance. You’ll explore how choices like chunking and query processing directly impact system performance, and how to move from simple retrieval to controlled, judgment-driven systems using metadata and reranking. The course also covers how retrieval systems fail in practice (such as returning plausible but incorrect results) and how to make these failures observable, through metrics and diagnostics. What makes this course unique is its focus on mental models over tooling. Rather than teaching a specific stack, it equips you with the concepts needed to understand, debug, and improve any retrieval-based system. By the end, you will be able to reason about retrieval behavior and make better design decisions in real-world Gen AI systems.