*We customize the course outline and content to your specific needs and relevant use cases.
Module 1: Designing maintainable LangGraph architectures
- Moving from small graphs to structured agent applications with clear responsibilities
- Separating orchestration logic, model interaction, tools, data access, and business logic
- Designing graph boundaries around reusable and independently understandable capabilities
- Choosing between deterministic workflows, agent driven decisions, and hybrid patterns
Module 2: Advanced state management and reducers
- Designing typed state schemas for messages, workflow data, intermediate results, and control information
- Understanding how reducers determine whether state values are replaced, combined, or accumulated
- Managing state updates safely when several nodes contribute to shared information
- Keeping transient execution data separate from information that needs to persist
Module 3: Advanced routing with Command and dynamic execution
- Combining state updates and routing decisions with Command
- Designing loops, retries, fallback routes, and conditional termination
- Using dynamic routing when the next step depends on model output or runtime state
- Preventing uncontrolled loops and keeping execution paths understandable
Module 4: Subgraphs and modular workflow composition
- Encapsulating reusable workflows as subgraphs
- Sharing or transforming state between parent graphs and subgraphs
- Using subgraphs to separate specialist capabilities and business processes
- Designing graph composition that supports independent development and testing
Module 5: Checkpointing and durable execution
- Understanding checkpoints, threads, and persisted graph state
- Configuring persistence so workflows can continue across multiple interactions
- Resuming interrupted or failed executions from stored state
- Designing idempotent node behavior when execution may restart or repeat
Module 6: Human-in-the-loop workflows with interrupts
- Pausing execution dynamically with interrupts when human input is required
- Reviewing, approving, rejecting, or modifying generated actions before execution
- Resuming graphs with updated information while preserving workflow state
- Designing approval points for sensitive or high impact actions
Module 7: Short term state and long term memory
- Distinguishing thread based conversation state from information stored across conversations
- Designing memory around user preferences, application facts, and reusable context
- Deciding what information should be persisted, retrieved, updated, or forgotten
- Avoiding unnecessary context growth and keeping memory relevant to agent behavior
Module 8: Advanced tools and multi-agent coordination
- Designing tool interfaces with clear inputs, outputs, errors, and side effect boundaries
- Coordinating specialist agents through routing, supervisors, or handoff patterns
- Sharing only the state and context required by each agent
- Choosing between one capable agent, several specialist agents, and deterministic subworkflows
Module 9: Parallel execution, streaming, and responsive workflows
- Running independent graph branches concurrently where tasks do not depend on each other
- Combining results from parallel branches through appropriate state updates
- Streaming model output and graph events to applications while execution continues
- Designing responsive workflows without making execution unnecessarily complex
Module 10: Error handling, retries, and reliable side effects
- Distinguishing transient model, tool, network, and application failures
- Applying retry and fallback behavior at appropriate workflow boundaries
- Designing idempotent external actions to avoid duplicate side effects after retries or resumes
- Managing timeouts, partial failures, and recoverable workflow states
Module 11: Testing, debugging, and LangSmith tracing
- Testing nodes, routing decisions, state transformations, and subgraphs independently
- Creating predictable tests for successful paths, edge cases, and failure conditions
- Using LangSmith traces to inspect model calls, tool execution, routing, and graph state
- Diagnosing unexpected behavior by separating orchestration problems from model and tool problems
Module 12: Production architecture and operational readiness
- Separating configuration, secrets, models, persistence, and external services across environments
- Designing graphs for maintainability, observability, controlled changes, and version evolution
- Reviewing security, permissions, tool access, resource usage, and operational ownership
- Building a practical production readiness checklist for future LangGraph implementations