RAG & Working with Your Own Data

by bam

Make AI answer over YOUR content (docs, policies, a product catalog, a knowledge base) without retraining a model. This is the technical, builder-track course for people who ship Retrieval-Augmented Generation (RAG). You'll build the whole pipeline: ingest and chunk your documents, embed them into vectors, store them (including in pgvector, which this app uses), retrieve the relevant pieces at question time, and assemble a grounded prompt. Then the parts that separate a demo from production: chunking strategy, retrieval quality (top-k, hybrid keyword+vector, re-ranking), context windows, and how to EVALUATE a RAG system: retrieval quality separately from generation quality, and groundedness above all. We cover the real failure modes (irrelevant chunks, stale data, prompt injection hidden inside documents), freshness and access control on your vector store, and the honest question most tutorials skip: when NOT to use RAG at all (RAG vs. fine-tuning vs. just-use-a-long-context-window). Carries the Learn.WitUS trust DNA throughout: a retrieved passage is evidence to ground and cite, never a fact to trust. F2 (Building with AI) is the recommended prerequisite; B2 (Agents & Automations) pairs well.