A model is only as useful as what it knows. This module teaches you to connect LLMs to real, trustworthy, up-to-date knowledge — your documents, your data, your world — using retrieval-augmented generation done properly.
Hallucination is what happens when a model has to guess. Retrieval-augmented generation replaces guessing with looking things up: the model answers from documents you control, and you can trace every claim back to its source.
This is the single most in-demand pattern in applied AI today — and you will build it end to end, by hand, before using any framework shortcuts.
The full pipeline: chunking documents well, embedding them, retrieving the right passages and grounding the model's answer — with the failure modes at each step.
Hands-on work with Pinecone, Weaviate, ChromaDB, FAISS and pgvector — what each is good at, and how to choose one for a real project instead of by fashion.
Why semantic search alone is not enough: combining vector similarity with keyword search and re-ranking so retrieval actually finds what matters.
Going beyond flat documents: building knowledge graphs that capture how facts relate, and querying them when questions need reasoning across sources.
Making AI remember across sessions with tools like Graphiti, mem0 and Zep — what to store, what to forget, and how memory changes an application's design.
How to know your RAG system is actually good: measuring retrieval quality and answer faithfulness, so improvement is engineering rather than vibes.
RAG is the skill employers ask for by name. Learn it end to end, free, with mentors beside you.