Engineering Reliable LLM Applications with LangGraph, MCP, and AWS

A documentation agent spots a change, calls a tool, and receives conflicting information. Does it publish, retry, or ask someone to review?

In this hands-on workshop, Anas Riad teaches how to build reliable AI agents through practical exercises you complete alongside him. You'll learn how LangGraph preserves progress, MCP connects tools, and validation, recovery, and human approval keep updates controlled.

Diagnose routing errors and invalid responses independently in focused Colab exercises, then review fixes together. Explore AWS deployment and monitoring through a prepared demonstration, with RAG and testing supporting the system.

Leave understanding how these pieces work together to keep an agent reliable when something goes wrong.

###

**This live event is for you because**

• You are an ML/AI engineer or LLM engineer developing applications that need reliable state management, tool use, and controlled execution.
• You understand basic LLM concepts and want to connect them within an agent system.
• You want to diagnose routing failures, invalid tool outputs, uncertain results, and retrieval issues.
• You want a practical understanding of permission boundaries, human oversight, and operational visibility.

**What you will learn and how you can apply it**

• Trace LangGraph state, conditional routing, checkpoints, and recovery paths to understand and correct workflow behaviour.
• Inspect MCP tool contracts and maintain clear boundaries between approved source reads and documentation writes.
• Validate structured outputs with Pydantic and handle invalid responses before they affect later decisions.
• Route uncertain or conflicting updates to human approval and verify selected behaviours with focused tests.
• Examine how retrieval over approved documentation supports answers with source references.
• Follow the AWS deployment path and use structured logs to investigate successful and failed runs.

**What you will build**

• Complete three Colab exercises to fix routing, invalid tool outputs, and human approval paths.
• Explore a prepared documentation agent that detects source changes and proposes updates.
• Examine read-only source access, restricted writes, and human review of uncertain changes.
• Inspect source-backed RAG answers and diagnose retrieval issues.
• Follow an AWS deployment demonstration and inspect application logs without deploying individually.

**Prerequisites**

• Basic understanding of LLM concepts, applications, and tool calling.
• Relevant Python, ML/AI, or LLM engineering experience, with enough Python proficiency to read, modify, and run short code examples.
• Familiarity with APIs and Git.
• No prior LangGraph or MCP experience is required.
• Basic Docker or AWS knowledge is helpful but not required.

**What you will walk away with**

• A practical understanding of stateful workflows, MCP tool integration, validation, recovery, and human oversight.
• Experience diagnosing and fixing selected project failures through focused exercises.
• A reference architecture you can adapt to documentation, support, research, and internal knowledge workflows.
• An understanding of the application's AWS deployment path and how structured logs support troubleshooting.
• Access to selected workshop resources, including project code, an architecture overview, and references for LangGraph, MCP, and AWS.