The problem that launched Michael Amori‘s company wasn’t glamorous — it was the mundane reality that even the world’s most advanced military struggles with broken-down equipment, misplaced parts, and maintenance schedules that seem designed by committee.
Nine years ago, when Amori and his team emerged from research labs at Caltech and NASA’s Jet Propulsion Laboratory, they saw an opportunity in what others might consider the unattractive side of defense technology.
While much of the AI industry chases autonomous weapons and sci-fi applications, Amori’s Virtualitics has built its business around a more fundamental challenge: keeping America’s military ready to fight.
The company’s focus addresses what may be the Pentagon’s most pressing technological need — not whether machines can replace humans in combat, but whether artificial intelligence can finally help humans make sense of the mountains of maintenance logs, supply chains, and personnel data that determine whether forces can actually deploy when needed.
Are you building game-changing tech for defense? 💪
We’re always looking to feature innovators pushing the boundaries in autonomy, AI, energy, and beyond. If your company belongs in the Defense Disruptors spotlight, get in touch with us here.
A Tight Focus on Readiness
Amori said Virtualitics is “entirely focused on delivering AI applications that increase the readiness of our defense customers.” By readiness, he added, he means the intersection of “maintenance, logistics, personnel, things that are key to making sure that the units that make up our military are ready to complete their mission.”
He offered aircraft maintenance as a straightforward example. According to Amori, Virtualitics’ software “is able to forecast the risk that certain components within the airplane are going to fail, and then recommend actions that are needed to avoid that specific part from failing,” while giving maintainers visibility into “the best next thing to do” amid competing requirements.
The company, Amori noted, serves exclusively defense clients and “came out of research done at Caltech and NASA JPL,” operating for “about nine years now.”
He described the firm’s identity in stark terms: “We think of ourselves as the mission AI company.”
What the Software Looks Like
Rather than one-off builds, Amori described a common framework with light configuration.
“We have built certain foundational software elements,” he said, applications for use cases like maintenance or material storage. Different solutions or platforms “still use the same exact framework,” he noted, with adjustments for how each customer’s data is structured or displayed.
Amori explained that the framework’s universality surprised even his team. “We’re even finding that whether it’s a ship or an airplane or a tank that needs to be maintained, there are a lot of commonalities in terms of how you approach it,” he said.
Access, he said, happens through Virtualitics’ own application. Users “log into our application, and they have a dashboard which basically gives them all the results from the AI.”
Under the hood, he added, the platform integrates with “other data providers,” aggregating underlying data into one place.
Why Explainability Matters
Amori repeatedly stressed explainability as a core differentiator. “AI is great, but it’s only great if it’s really understandable by the people using it,” he argued, claiming that opacity creates “a huge implementation barrier.”
He said the company addresses this challenge “from a user interface and user experience point of view,” as well as through “algorithm design” and the “output that comes out of all the algorithms.”
The emphasis on explainability, Amori suggested, stems from practical necessity in military environments where users need to understand why systems recommend specific actions.
He also pointed to in-house domain expertise as crucial. About “22 percent of our employees are veterans,” he said, including former Air Force maintainers.
In his view, that experience helps ensure products “meet the needs and are explainable enough” while teams working with customers can “make sure that the value really shows.”
Wrestling With Data — and the Legacy Stack
Asked about integrating with complex, aging military systems, Amori acknowledged the sprawl. “There are a lot of legacy systems,” he said, with data “located in many different places” and often containing “a lot of noise” and “issues with the quality of the data.”
He attributed much of the company’s engineering effort to accessing, merging, and cleaning that information.
“Our engineers are very good at aggregating the data from different sources,” he said, but emphasized that domain knowledge “is of crucial importance” because teams must understand what maintenance data spread across multiple databases actually means.
“The old adage is garbage in, garbage out,” Amori noted. “So you’ve got to make sure that what you’re inputting in the algorithm makes sense. It’s clean, it’s well-structured, and getting there is difficult.”
Security and Where the Software Lives
On security, Amori emphasized that Virtualitics does not host government data. “Once we deploy the application to the customer, we don’t host the application itself, so we are not privy to the data,” he explained. “We deploy on the customer’s cloud.”
Specifically, he added, deployment happens “on DoD cloud networks, not on our own network.”
To enable such deployment, the software must pass the Pentagon’s Authorization to Operate process, which “involves making sure that we meet a bunch of different controls within the company,” according to Amori.
Measuring Impact
When discussing a recent US Air Force contract, Amori declined to share operational details but emphasized the company’s focus on measurable outcomes.
“We really care about our customers,” he said, describing Virtualitics as a company of “about 100 people” that must balance financial sustainability with mission impact.
The company tracks success, he explained, by working with customers to define key performance indicators relevant to their operations, then monitoring whether the AI applications actually move those metrics.
He offered parts availability as one example, suggesting that effective predictive maintenance should reduce situations where “things break down without us knowing about it” and create parts shortages.
“We spend a lot of time with the customer defining what those KPIs are, and then tracking those KPIs to make sure that what we’re delivering is really doing something meaningful,” Amori said.
Visualizing the ‘Why’
Amori said the firm employs advanced visualization techniques — including “3D immersive visualizations, kind of like a video game” — to help users understand model outputs. Such approaches, he argued, “really improve the human’s ability to understand what it is that the AI is trying to tell.”
He also highlighted the company’s use of knowledge graphs, which he described as “part of our patented technology.” The approach allows teams to “take unstructured data and visualize it as a network” to show “connections between the different data points.”
The visualization strategy, Amori suggested, serves the broader goal of explainability: “We always want to make sure that the human understands why the AI is recommending what it’s recommending.”
Where AI Helps Most
Looking ahead, Amori positioned readiness as both the biggest opportunity and the area where current AI capabilities align best with military needs. He called readiness “the number one priority” within the Defense Department and “the largest budget item within the DoD.”
The problems involved — predictive maintenance, supply storage, personnel training requirements — represent areas where “AI, as it currently stands, is well suited to provide some very good assistance to the human,” according to Amori.
“We’re just scratching the surface,” he said of the company’s current work, adding that Virtualitics plans to “zoom in on readiness” because that’s where both demand and technological capability converge.
The company’s trajectory, Amori suggested, reflects a broader theme in military AI adoption: the most impactful applications may be those that enhance human decision-making in mundane but critical tasks, rather than replacing human judgment altogether.
