https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/Recent content in Retrieval-Augmented Agents on DATATWEETSHugoenCopyright (c) 2025 DatatweetsSun, 28 Jun 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-1-why-agents-need-retrieval/Fri, 23 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-1-why-agents-need-retrieval/An agent answers from frozen training knowledge: it can’t see your private docs or recent data, and it fills gaps with confident guesses. Retrieval grounds the agent in a knowledge base you control. Meet the retrieve-augment-generate pattern that the rest of this module builds.https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-2-building-a-knowledge-base/Fri, 23 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-2-building-a-knowledge-base/A knowledge base is the memory module’s machinery pointed at documents: chunk each document into passages, embed them, and rank them against a query by cosine similarity. Build a small KnowledgeBase with add_document and search, run it dependency-free on two travel guides, then swap embed() for real semantic embeddings in production.https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-3-retrieval-as-a-tool/Fri, 23 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-3-retrieval-as-a-tool/There are two ways to do RAG: always retrieve before answering (a fixed pipeline), or give the agent a search_knowledge tool and let it decide when to retrieve. This lesson builds the tool, wires it into the agent loop, and walks a verified trace where the agent chooses to look something up and cites the source.https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-4-grounding-and-citations/Fri, 23 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-4-grounding-and-citations/Retrieval gives the agent sources, but only discipline makes it trustworthy. You’ll build a grounding gate that refuses out-of-knowledge questions before the model is ever called, prompt the model to answer only from numbered sources with inline citations, and tune a similarity floor that cleanly separates real matches from keyword noise.https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-5-guided-project-retrieval-augmented-atlas/Fri, 23 Jan 2026 09:00:00 +0200https://datatweets.com/courses/ai-agents/retrieval-augmented-agents/lesson-5-guided-project-retrieval-augmented-atlas/Put Module 6 together on one agent: build Atlas a destination knowledge base, give it a search_knowledge tool to retrieve mid-plan, ground its answers in retrieved passages with citations, and refuse honestly when the knowledge base has nothing. Then tie in memory so Atlas plans from both what it remembers about the traveler and the facts it looks up.