Most monitoring systems can alert when something looks wrong, but very few can say why. Engineer in a Box, Keyfive’s AI for explaining equipment anomalies, is built to answer that question.
It identifies changes in performance, compares them with knowledge of how that specific piece of equipment should behave, and offers an easy-to-understand explanation of what likely happened.
“We’re proceeding towards an entirely unsupervised, non-expert AI system that’s meant to make it so that anyone can understand what’s going on with their system,” says Daniel Barnett, CTO of Keyfive.
Why anomaly detection alone leaves teams guessing
AI is already changing how the industry identifies and responds to problems and anomalies. According to the International Energy Agency (IEA), “AI-based fault detection can help rapidly identify and precisely pinpoint grid faults, reducing outage durations by 30-50%.”
But knowing that something is wrong, and even where, isn’t the same as knowing why. Most anomaly detection stops at the alert and leaves root cause analysis (RCA) to the experts. In critical operations, this can mean calling in staff overnight and tasking highly skilled people with manual data analysis to get systems back online.
Those experts are becoming harder to find. The IEA World Energy Employment Report 2025 found 2.4 energy workers nearing retirement in advanced economies for every new entrant under 25. When the engineer who understands a site’s equipment is busy or no longer on the team, anomalies sit unexplained.
How Engineer in a Box finds the likely root cause
With Engineer in a Box, anomalies are put into context. It begins by building a normal performance profile of the equipment, using the manufacturer’s specifications and the device’s operating history.
“With this, Engineer in a Box knows what’s normal for this device and what to expect,” Barnett says.
The anomaly readings are compared to the profile. Statistical models then determine whether the change is significant and whether it affects the system’s health, safety, or performance.
Engineer in a Box offers a plain-English summary of what happened and what needs attention, as well as a technical breakdown for engineers. Teams can add their own findings to the record, giving each explanation more context.
AI that keeps learning from your equipment
When operators confirm results or provide additional context or maintenance records, the labeled history is committed to memory and makes the next explanation more accurate.
As operations grow, baseline models are established across fleets and deployments of similar equipment, so each device can be compared with others like it.
Engineer in a Box also draws on the rest of Keyfive’s monitoring and reporting tools. The Performance Scorecard provides a month-by-month record of how each site performs, and Fast Telemetry captures high-resolution data from the moments around an event.
Getting ahead of equipment problems
Engineer in a Box can’t predict failures, but it can help teams act sooner. Many equipment problems start as small changes long before they cause an outage. By flagging early signs of degradation and trends heading in the wrong direction, Engineer in a Box gives teams time to plan maintenance in ways that reduce both cost and disruption.
Industry research supports that approach. A study of battery storage failures by EPRI, PNNL, and TWAICE found that most failures came from components other than the battery cells and called for better monitoring and analytics during operation to spot potential failures earlier.
“Our solution provides a unified view from an intelligent AI agent that tells you what’s going on with the device, factoring in all of its history, specifications, and data streams,” Barnett says.
Building AI the energy industry can trust
The IEA’s Energy and AI report found that AI skills are much less common in energy than in other sectors. For AI to earn a place in operations, it has to be easy to understand and grounded in data teams can verify.
“Everything in the base layer of our solution is machine learning, and then the agent is added on top for context,” Barnett says. “Our summaries come from primary data sources. This is very much a trust-oriented AI system.”
That foundation is what lets Keyfive move toward more automation without giving up trust.
Next, Keyfive plans to have the edge device score unusual events and send them straight to Engineer in a Box, so alerts arrive with an explanation already attached and teams can act right away.
Learn more about Keyfive’s latest technology releases in The Current.
FAQ
Is Engineer in a Box predictive maintenance?
No. It explains events that have already happened and can flag trends heading in the wrong direction, but it doesn’t predict failures.Does it need training data or a data scientist to set up?
No. The anomaly detection is unsupervised, so it learns what’s normal for each device from its own data, without manual labeling first.What equipment does it work with?
It builds profiles for power system devices like generators and battery systems, utilizing available performance history and manufacturer specifications.Does it replace an engineer?
No. It does the first round of analysis so engineers can spend their time on the problems that need them.
Learn how Keyfive’s Engineer in a Box can make your systems more intelligent.