Introduction to LangGraph (LangChain Academy)

Topics covered

• •Set up the working environment: Python 3.11+, a virtual environment, dependencies, API keys.
• •Understand the unified chat model interface in LangChain: `invoke`, `stream`, and the `model` and `temperature` parameters.
• •Learn to work with messages (`HumanMessage`, `AIMessage`) and connect the Tavily web search tool.

Topics covered

• •Build your first graph: state (`TypedDict`), node functions, plain and conditional edges, `START`/`END`.
• •Wire an LLM into the graph: a chain with tool calling and a router.
• •Assemble a full ReAct agent with a "model → tools → model" loop and add memory via a checkpointer.

Topics covered

• •Learn different ways to describe a state schema: `TypedDict`, `dataclass`, Pydantic — and when you need runtime validation.
• •Understand reducers: why they're needed for parallel updates and how to write a custom one.
• •Learn to separate schemas: private state between nodes, and separate input/output schemas for a graph.

Topics covered

• •Learn to stream graph execution: `values` and `updates` modes, and token-by-token streaming via `astream_events`.
• •Set breakpoints (`interrupt_before`) so a human can approve agent actions.
• •Edit the state of an interrupted graph: `update_state`, a stub node for human feedback, `as_node`.

Topics covered

• •Run nodes in parallel (fan-out/fan-in) and understand why shared keys require a reducer.
• •Move logic into subgraphs with their own state schemas and connect them via shared keys.
• •Apply map-reduce via the `Send` API: a dynamic number of parallel branches.

Topics covered

• •Distinguish short-term memory (within a thread, checkpointer) from long-term memory (across threads, Store).
• •Learn the LangGraph Store: `namespace`, `key`, `value`, and the `put` / `get` / `search` methods.
• •Save memory against a schema: a user profile via `with_structured_output` and its limitations.

Topics covered

• •Understand the components of the LangGraph Platform: LangGraph Server (API), Redis, Postgres, CLI, SDK, Studio.
• •Build a deployment for the `task_maistro` app: `langgraph.json`, a Docker image, a `docker-compose` with three services.
• •Connect to the deployment: the LangGraph SDK and Remote Graph; run background and blocking runs.