Case Study

Major U.S. Utility Installs a Semantic Layer for Its AI Agents on AWS

At a glance

A major U.S. electric utility could not quickly answer three questions every security and audit team asks: which team runs this system, what a compromised account can reach, and who was responsible on a given date. Every minute spent digging for those answers added to incident response time, the number the security team works hardest to keep low.

New Math Data set up an ontology and knowledge graph on AWS with the utility’s team, built on open standards. It reads organizational data from the utility’s identity directory without writing back, and links it to dated system ownership records, so one query answers ownership, escalation and access-reach questions. The identity directory, the systems and the team’s workflow stayed the same.

The utility now has one governed map of people, teams and systems that its AI agents and analysts query in plain business terms, and a pattern each new system plugs into.

Industry

Use Case

Agentic AI Infrastructure, Knowledge Management, Security & Monitoring, Compliance & Governance, Semantic Layer

Solution implemented

The value equation

Company Snapshot

The client is a major U.S. electric utility operating a multi-state grid for millions of residential and commercial customers.

Location

US

Customer Situation

Ownership and access data lived in the identity directory, group naming conventions, application owner records and team knowledge. No two sources named a person, team or system the same way, so answering one ownership or access-reach question meant manual digging across all of them. There was no single triggering event. The team saw it as a gap AI could close, as long as the approach met three conditions: read-only access to identity data, open standards instead of a proprietary model, and no new system of record.

Many large enterprises share this problem: identity and IT ownership sit in silos, so security, audit and AI governance questions take days instead of seconds.

NMD Solution

NMD reviewed the utility’s sources and found that the data existed, but no shared vocabulary connected it.

  • Model: a layered ontology built on public standards (W3C ORG, FOAF, PROV-O, ITSMO, SKOS), plus a thin responsibility extension where every owner and escalation contact carries start and end dates.
  • Extraction: an MCP (Model Context Protocol) server with typed, read-only identity directory tools that pull users, groups, service principals, applications and owner assignments, mapped onto the ontology.
  • Knowledge graph: the directory extract plus responsibility records for a data warehouse and a web application, held in an RDF/SPARQL store on AWS.
  • What stayed the same: the identity directory remains the source of truth, and nothing is written back to it.

What We Delivered

Security and IT teams ask one question and get the owner, escalation contact and access reach of a system, including who was responsible on any past date. The utility’s team received the runbook, code and training to run it. Agents use the same governed tools, so each new system plugs into the existing vocabulary instead of starting a new project. The same model extends next to more systems and to grid asset data.

Ready to Transform Your Utility Business?

See how New Math Data can transform your utility business with AWS-powered innovation.