A US fashion retailer’s procurement agent hallucinated stock — complete with fake SKUs and fabricated inventory levels. A global bank’s infrastructure resilience agent invented servers and data centres that didn’t exist. Different domains, identical failure mode: LLMs confidently generating things that aren’t real.

Max Latey (Pinboard Consulting) walks through the GraphRAG architectural patterns his team has deployed across radically different industries — fashion retail, FMCG, law firms, and financial services — and shows why the same core approach works in all of them.

The session covers what stays the same across domains: schema-first ontology design, multi-agent query generation, and mixtures of pre-canned and agent-originated queries. And what changes: domain-specific entity types, validation rules, and data integration patterns.

Max will show how discovered schemas become fixed ontologies, why iterative query generation with a validation agent outperforms single-shot querying, and how n8n-based multi-agent workflows handle the full loop from question to grounded answer.

If you’re building GraphRAG and deciding whether to let your schema drift or lock it down, this session will save you the wrong choice.

Unique Domains, Common Solutions - Killing AI Hallucinations from Fashion Retail to Banking Infrastructure