Agentic AI
Advanced
Advanced LangGraph Workflows
Design durable stateful workflows for tool-using AI applications.
Course Overview
Explore graph state, checkpoints, branching, and failure recovery through a production-shaped workflow. The course focuses on control flow and observability rather than generated recommendations.
What You Will Learn
Design stateful, multi-step graphs using LangGraph state reducers
Implement durable check-pointing for pause, resume, and human-in-the-loop steps
Build parallel branch execution and conditional router nodes
Handle transient node failures with backoff retry policies and safe rollback states
Monitor graph execution flows using telemetry and node execution metadata
Tools & Technologies Used
LangGraph
Python 3.11
Asyncio
Sqlite/PostgreSQL
LangChain
Structured Curriculum
2 Modules · 6 Lessons · 240 Minutes Total
Module 1
Module 1: LangGraph Architecture & State Machines
3 lessons
Module 1: LangGraph Architecture & State Machines
Master graph state definitions, custom node reducers, and flow control.
-
State Reducers and Graph Topologies
Understand state immutability, state mutation keys, and reducer mechanics.
-
Constructing Conditional Routing Nodes
Build routing functions that evaluate confidence scores to branch execution paths.
-
Parallel Branch Merging & Fan-Out
Execute parallel retrieval and evaluation nodes before aggregating results into a final state.
Module 2
Module 2: Durability, Checkpointing & Fault Tolerance
3 lessons
Module 2: Durability, Checkpointing & Fault Tolerance
Integrate persistent state checkpoints for fault recovery and manual approvals.
-
Database Checkpointers in Production
Configure SQLite and Async SQLAlchemy checkpointers to persist thread state across restarts.
-
Implementing Human-in-the-Loop Interrupts
Pause graph execution for human verification before high-risk database mutation steps.
-
Graph Recovery & Retry Policies
Implement sub-graph error boundary traps and deterministic fallback nodes.
Practical Project & Capstone Outcome
🚀 Capstone Project
Self-Correcting Multi-Step Code Refactoring Graph
Design a 5-node LangGraph workflow that accepts code, executes static linting, identifies issues, invokes an LLM repair node, and re-tests until clean or max retries are reached.
Prerequisites
- Completion of 'Introduction to Agentic AI' or equivalent experience
- Strong understanding of async Python (`asyncio`) and state machines
Intended Audience
- Senior Software Engineers building complex LLM orchestration pipelines
- Data Engineers orchestrating multi-stage AI workflows
Instructor Information
J
Jon Bell
Course Author & Industry Expert
Jon Bell is a Lead Distributed Systems Engineer specializing in stateful workflow engine design. He has authored several production graph frameworks.
Frequently Asked Questions
Is LangGraph required for stateful agents, or can I use raw Python?
While raw Python works for trivial loops, LangGraph provides production-grade state persistence, human interrupts, and branching control.
Does this cover version 0.2+ of LangGraph?
Yes, all lessons use modern LangGraph syntax with typed state dictionary reducers.