LangGraph

Status

✅ Notes complete — 13 subtopics distilled from the CampusX LangGraph course (~595 pp), official docs, and the CampusX GitHub repos.

Reading order

Read in order — each topic compounds on the previous. Chapters 1–3 are foundations; 4–7 are workflow shapes; 8–10 are durability/UX; 11–13 are real applications.

#	Topic	Read
1	Introduction — Agentic AI, LangGraph vs LangChain	01-introduction
2	State & Schema — TypedDict, Annotated, Reducers	02-state-and-schema
3	Graph Basics — Nodes, Edges, Compile	03-graph-basics
4	Sequential Workflows	04-sequential-workflows
5	Conditional Workflows	05-conditional-workflows
6	Iterative Workflows — Loops & Cycles	06-iterative-workflows
7	Parallel Workflows & Fan-Out	07-parallel-workflows
8	Persistence & Checkpoints	08-persistence-and-checkpoints
9	Streaming	09-streaming
10	Human-in-the-Loop	10-human-in-the-loop
11	Tools & the ReAct Agent	11-tools-and-react-agent
12	Subgraphs & Multi-Agent	12-subgraphs-and-multi-agent
13	Real-World Apps — Chatbot, RAG Agent, MCP	13-real-world-apps

The big picture

flowchart TB
    subgraph Foundations [Foundations 1-3]
      A1[State Schema] --> A2[Nodes & Edges]
    end
    subgraph Shapes [Workflow Shapes 4-7]
      B1[Sequential] --> B2[Conditional] --> B3[Iterative] --> B4[Parallel]
    end
    subgraph Durability [Durability + UX 8-10]
      C1[Persistence] --> C2[Streaming] --> C3[HITL]
    end
    subgraph Apps [Real Apps 11-13]
      D1[Tools + Agent] --> D2[Multi-Agent] --> D3[Chatbot / RAG / MCP]
    end
    A2 -.-> B1
    B4 -.-> C1
    C3 -.-> D1

Learning roadmap

• Why a graph — limits of linear chains
• Core concepts — StateGraph, nodes, edges, state schema
• Reducers — add_messages, operator.add, custom
• Sequential / Conditional / Iterative / Parallel workflows
• Cycles, loops, and recursion_limit
• Send API for dynamic fan-out
• Checkpointing & persistence (MemorySaver, SqliteSaver, PostgresSaver)
• Streaming modes (values, updates, messages)
• Human-in-the-loop interrupts (static & dynamic)
• create_react_agent & ToolNode
• Subgraphs & supervisor multi-agent patterns
• Chatbot / agentic RAG / MCP client blueprints

What I should be able to do after this topic

• Decide when to use LangGraph vs LangChain LCEL vs plain Python.
• Design any multi-step or multi-agent workflow as a graph.
• Add checkpointing so a long-running agent can resume.
• Insert human approval steps before risky actions.
• Stream intermediate state to a chat UI.
• Build a production-ready agent that uses tools / retrieval / MCP.