Intermediate Overview
What Intermediate covers, what you'll build by the end, and how it builds on the Q&A bot you already made in Foundations.
Chapter 1: Chunking Strategies
Why chunk size matters for RAG, and how fixed-size, recursive, and semantic chunking each cut a document differently.
Chapter 2: Choosing an Embedding Model
Weighing OpenAI against open-source embedding models on quality, latency, and cost, instead of just accepting whatever model a tutorial happens to use.
Chapter 3: Better Retrieval
Why vector search alone gets the wrong chunk more often than you'd think, and three ways to fix it: metadata filtering, hybrid search, and re-ranking.
Chapter 4: Prompt Patterns
Three prompt patterns that fix three failure modes: chain-of-thought for wrong shortcut answers, structured/JSON output for unparseable replies, and function calling for decisions the model can't act on.
Chapter 5: Tool Use
Two real tools, a calculator and a Wikipedia search, wired into a loop that actually runs what the model decides and feeds the result back into the conversation.
Chapter 6: Your First Agent
The same calculator-and-Wikipedia agent from Chapter 5, rebuilt with LangChain's create_agent, so you can see exactly what a framework does for you and what it hides.
Chapter 7: Memory
Give your Chapter 6 agent memory across turns, first the simple way (remember everything), then the scalable way (summarize the old parts once the conversation gets long).
Chapter 8: Evaluating What You Built
Stop eyeballing whether your RAG pipeline works. Measure retrieval with precision/recall, and measure generated answers with a second LLM acting as judge.
Chapter 9: Capstone — Multi-Tool Agent
The Intermediate capstone — one agent, three tools it picks between on its own: a calculator, Wikipedia search, and RAG over your own documents, all with memory across the conversation.