
Opening keynote: Six success elements in industrial AI
Plant data needs a knowledge graph, not a bigger model.
Moe Tanabian, Founder and CEO, IntuigenceAI
- Onboarding no longer takes a year. Building a knowledge graph for a typical plant of about 500 assets used to take nine to eighteen months. With AI reading the P&IDs, PFDs, time series and tens of thousands of documents, a working graph now takes days.
- Zero egress skips the longest approval cycle. All 17 components, the AI models included, are licensed into the customer’s own Azure tenant. Nothing leaves, so the three to six months of CIO, CISO and legal review that egress triggers never starts.
- Frontier models fall off a modality cliff on plant data, and charge more for it. Models that score 80 to 90 percent on text answered fewer than 10 percent of connectivity questions across 20 P&IDs. Domain-trained smaller models answer at 10 to 100 times lower cost, and a graph answers in milliseconds where a model reasons for minutes.
“If you want to build a demo, build it with a large language model. If you want to give it to 50 engineers in the field, build a knowledge graph.”Moe Tanabian










