AI agents that book 15 guests in a 10-person room. Agents that fabricate statistics when data doesn't exist. Agents that pick wrong tools from 29 options, wasting $47 in tokens. These aren't prompt engineering failures - they're architectural limitations that need structural solutions. This hands-on workshop covers 5 research-backed techniques to prevent agent hallucinations: 1. Graph-RAG (Neo4j) - Replace vector similarity guessing with precise entity relationships. Result: 73% fewer fabricated statistics. 2. Semantic Tool Selection - Filter 29 tools to the relevant 5 using embeddings. Result: 89% token reduction, accurate tool selection. 3. Multi-Agent Validation - Executor-Validator-Critic swarms catch fabrications through cross-checking. Result: 92% detection rate. 4. Neurosymbolic Guardrails - Framework-enforced rules (lifecycle hooks) that agents cannot bypass. Result: Zero business rule violations. 5. Agent Steering - Guide agents to self-correct instead of blocking them. Result: Task completion without hard failures. Each demo includes live code, before/after metrics, and failure case analysis. Final module shows production deployment with DynamoDB and Lambda. You'll walk away with working Python implementations, a decision framework for when to apply each technique, and an open-source repository adaptable to your domain. No AWS account or cloud costs required - sandbox environments provided for all hands-on exercises. Basic Python experience and LLM familiarity recommended but not required.

Outline: • Introduction - Why AI Agents Hallucinate Differently Than LLMs • Demo 00 - Strands Agents Primer • Demo 01 - Graph-RAG vs. Standard RAG • Demo 02 - Semantic Tool Selection • Demo 03 - Multi-Agent Validation • Demo 04 - Neurosymbolic Guardrails • Demo 05 - Agent Control Steering • Demo 06 - Production on Amazon Bedrock AgentCore • Workshop Recap + Resources • Q&A