The course introduces AI agent design with LangChain: from agent architecture and tools, through memory and step orchestration, to testing patterns and safer deployment of agentic applications. Lessons will be added in a later stage.
Learn the core concepts needed to build agentic applications deliberately: what LangChain is, when it makes sense, how to distinguish chains, agents, and workflows, and how to prepare a safe project environment.
Learn the Runnable interface, basic execution modes, and how to compose simple LCEL chains: from prompt template, through model, to output parser and debugging.
Learn tools as controlled agent actions: how an agent chooses them, how to design custom tools, how to validate arguments, and how to safely handle errors.
This chapter moves from a first one-tool agent to multiple tools, the ReAct loop, and iteration and cost limits.
You will learn how to connect agents to documents: from a retriever as a tool, through chunking and embeddings, to source-grounded answers and hallucination control.
You will learn how to design output schemas, enforce format, return application-ready data, and handle retries after validation errors.
You will learn when a free-form agent becomes too brittle and how LangGraph helps design explicit stateful workflows with human control and checkpoints.
This chapter shows how to test prompts and tools, use LangSmith, read traces, and evaluate the quality of agent answers.
