Road to NODES 2026: Stop AI Agent Hallucinations: 5 Techniques + Production Patterns
Your agent books 15 guests into a 10-person room. It invents statistics when the data isn’t there. It picks the wrong tool out of 29. Better prompts won’t fix this, because the cause is architectural.
In this workshop, you implement five techniques in Python, each with live code and before-and-after metrics:
GraphRAG: ground answers in precise entity relationships instead of vector similarity guesses
Semantic tool selection: narrow large toolsets to the relevant few and cut token usage
Executor-Validator-Critic swarms: catch fabrications before they reach users
Neurosymbolic guardrails: enforce rules through lifecycle hooks agents can’t bypass
Agent steering: guide agents to self-correct instead of failing
You leave with a decision framework for choosing the right technique, a module on production deployment, and an open-source repo to adapt. No AWS account or cloud costs are needed, since sandboxes are provided.
Trainer: Elizabeth Fuentes Leone
Resources mentioned





