From a wall of alarms to a conversation
CORE manages thousands of connected lighting devices across enterprise buildings, controllers, sensors, wall stations, relay packs. At scale, the platform generated a relentless stream of alarms and status codes that only specialists could decode, and maintenance stayed stubbornly reactive: teams found out a device had failed only after it went dark.
My mandate was to define how AI should live inside an established enterprise product, not a bolted-on chatbot, but woven through the daily workflow. Three pillars emerged: conversational access, explainability for every alarm, and prediction ahead of failure.
Signal that only specialists could read
A single large site carried hundreds of devices and a daily flood of fault codes, almost none of which a facility manager could decode without help. AI had an obvious job from day one.
One bar to ask the building anything
A single Ask AI & Search bar sits atop every screen, with one-tap intents for the questions teams ask each morning. The same answer reads three ways, a floor plan, a building roll-up, or a table.
Explanation, plus a next step
AI only earns its place if it helps the person on the floor at 2am. Framing the assistant as explanation plus a concrete action, not open chat, kept every answer specific, sourced from a real device, and trustworthy.
Every alarm, translated to a plan
The shipped Explain Alarms panel pairs a plain-language cause with a numbered fix, while Predictive Maintenance ranks devices by silence and estimated lifespan so the worst actors surface first.
From a brief to the right fixture
The same intelligence reaches sales and specification. A conversational Fixture Recommendation assistant turns a plain brief into specific products, then connects to a searchable library of 5,000 IES files and an Area Wizard that auto-places fixtures on a real site map.
Searchable IES library, 5,000 photometric files filtered and exported in a click · Smart Layout Area Wizard auto-places up to 20 fixtures to hit a target level and uniformity.
Built for the people holding the pager
Anchoring AI to existing pain, the alarm wall, gave it an obvious job from day one. The next step is making confidence and provenance first-class, showing why the AI predicted a failure, not just that it did, and exposing what the optimization loop has learned over time.
What intelligence bought
The goal was never to add AI to a dashboard. It was to make a 400-device building answer a question, and tell you what to do next, the way the best technician on the team would.
