2027 Event Agenda

Modelers and engineers at Dominion Energy spend a meaningful share of their time not doing engineering, but hunting. Hunting through spreadsheets, enterprise planning documents, project documentation, and load projections they receive from different parties that don't align, trying to reconstruct context that should already be at their fingertips. The cost isn't just productivity. It's decision quality: assumptions go unvalidated, load projections and other critical pieces of information have no clear lineage. Institutional knowledge walks out the door, and onboarding new engineers into an undocumented system compounds the problem.

Dominion Energy and Data Society set out to fix this by building a planning and modeling portal with an embedded AI Copilot - a system that lets engineers ask questions in plain language and get answers grounded in actual planning data. But before a single query could be answered reliably, the foundational problem had to be solved: building a curated, reconciled data backbone where the lineage of every data point is clear, visible, and traceable from source to answer. The system combines retrieval-augmented generation (RAG), semantic search, and domain-specific knowledge grounding across enterprise planning documents, load projections, project records, and engineering analysis - all tied to an automated data pipeline that makes provenance a first-class feature, not an afterthought. A pilot is currently underway with a limited user group, with broader rollout planned ahead of the conference.

This session is a practitioner's account of what it actually took - and what we got wrong. We'll cover why data lineage is the unglamorous but non-debatable prerequisite that determines whether an AI planning assistant is trustworthy or just fast, how we curated and reconciled data sources that were never designed to coexist, why standard RAG approaches broke down against engineering documents, and how user trust becomes the real adoption bottleneck long before technical accuracy does. We'll also share how we evaluated the system and what the pilot taught us that the architecture phase didn't.

Attendees will leave with a grounded, jargon-free picture of how to responsibly introduce generative AI into utility engineering workflows - including the parts vendors don't put in their slide decks.

Learning Objectives:

• Understand why data lineage and source reconciliation are the prerequisite - not the afterthought - for trustworthy AI in utility planning environments.
• Learn how Dominion Energy and Data Society built a curated, automated data backbone that makes every answer traceable to its source.
• Explore where standard RAG implementations fail against engineering documents and what the architecture actually needs to handle it.
• Take away a practical framework for evaluating utility AI assistants on trust and transparency, not just accuracy.