AI in Business Intelligence

Topic Change Announcement! Serge will be doing his Survey of Latest AI Research next month.

New Topic: AI in Business Intelligence

LLMs are incredible at understanding natural language, but ask one what "yellowcake" means and you might get a dessert recipe instead of a uranium commodity price. Domain experts speak in shorthand, acronyms, and jargon that shifts meaning depending on context. "HA" means heart attack to a cardiologist, headache to a GP, and hazard assessment to a nuclear engineer. Fine-tuning is expensive, fragile, and opaque. There's a better way. In this talk, Bryan walk through a working system that pairs the probabilistic power of LLMs with deterministic domain definitions stored in a SKOS knowledge graph. Using the nuclear energy industry as a real-world example, he'll show how a SPARQL-backed glossary connected to an LLM via the Model Context Protocol (MCP) lets domain users ask questions in their natural language — "What's the current SWU price?" — and get the right answer every time, without fine-tuning or prompt engineering gymnastics.

About the speaker:
Bryan Corder is an Engineering Manager over the AI and Data Innovation department at Hagerty Insurance. Before going into Data Engineering, Bryan worked in commodities for a British investment firm and a Japanese trading company. He holds an M.S. in Computer Science from Franklin University and an M.A. in International Relations from American University.