GRAPEVINE, Texas — Data scientists and other technical specialists were embedded next to uniformed military personnel inside U.S. Transportation Command’s global operations center for the first time during Operation Epic Fury to rapidly prototype AI-enabled tools in support of real-world warfare needs.
Those experts generated roughly “33 apps in just the first two weeks alone of OEF,” Transcom’s Chief Data and AI Officer Ryan Samuelson told DefenseScoop.
In an exclusive interview at the National Defense Transportation Association’s annual Fall meeting in Texas last week, Samuelson shed new light on Transcom’s use of AI and the Maven Smart System to accelerate strategic decision-making and data-sharing over the course of the Iran war.
“Decision advantage is how you shape outcomes ahead of when you need to actually move,” he said.
‘The power of it’
Transcom is responsible for moving troops and cargo worldwide for all U.S. combatant commands and defense agencies, and also transferring sick and injured troops to medical facilities for specialized treatments and recovery.
The Joint Distribution Process Analysis Center — or JDPAC — operates as the command’s analytic, engineering, and modeling engine within the global defense transportation network.
Transcom is known as an influential early adopter of Palantir’s Maven Smart System in the U.S. military enterprise. That operational AI and data-fusion software platform grew out of the Defense Department’s Project Maven, which launched nearly a decade ago.
MSS pulls disparate real-time data streams — such as satellite imagery, sensor telemetry, radar, flight paths, ship trackers and classified intelligence — into a single interface and uses computer vision and automated detection models to identify, track, and categorize potential military targets and other assets of interest.
It’s built on Palantir’s Ontology, which the company describes as “an operational layer” that “sits on top of the digital assets integrated into” the platform (like datasets and models) and connects them to their real-world counterparts (like facilities and equipment).
“I don’t want to be particular to a certain vendor — but I will tell you, in the Maven Smart System Palantir Ontology — the reason JDPAC actually was the first in Transcom to use it was the power of it. What we realized is there’s not much else, at least right now, on the market that has the ontology to be able to put in the data, to be able to put in the apps, to have the secure development. It is the most powerful tool,” Samuelson said. “The ability to operate within that, to be able to share, is really, really powerful.”
According to the CDAO, when President Donald Trump initiated Operation Epic Fury in February, Transcom’s leaders instructed data scientists and engineers from JDPAC to work side-by-side with the command’s operators.
“So, instead of those folks being in a different building and having their own work processes that would get an email or something, we put them on the floor so that they could see on all the screens the maneuver elements, the force flow, and what was happening in real time with Iranian counterattacks and U.S. attacks,” Samuelson told DefenseScoop.
Transcom’s operational staff were directed to bring problems to the data and tech experts, who would then build prototypes to help solve them in real time as the conflict unfolded.
Samuelson said he tasked his team to deliver a minimum viable product within 24 hours, gather feedback, and continuously iterate to refine their tools. Everything was connected and deployed in MSS, and the team generated dozens of digital apps and dashboards.
Some of those prototypes are still in use today.
“What did that allow us to do? I’ll give you examples. And it goes all the way from our political advisor to units at echelon that could actually use the tools,” including other combatant commands, Samuelson said.
‘Work at the speed of war’
From the start of Epic Fury, Transcom’s political advisor had to make sense of what Samuelson said were “all these State Department cables that were flowing in.”
The tech embeds on the operations floor quickly built an AI tool that could scrape all of those cables and summarize the ones that would likely impact Transcom’s mission set, such as those related to contested logistics or access basing and overflight restrictions.
Instead of humans having to sift through hundreds of pages of material to do that as they were coming in, the bot was actively looking and generating its own reports and query responses.
“We ended up taking the tool and all of the data that we dumped in Maven Smart System — when planes are moving, what the cargo flows were, all of that — and over a 24-hour period, our bot would look at all of that. And then [Transcom Commander Gen. Randall Reed] could come in every morning and go, ‘What happened in the last 24 hours?’ And in a matter of about 15 seconds, it would spit out a summary of, ‘This is your movements. Here’s how many were late, et cetera, et cetera,’” Samuelson said. “That used to take a team of probably four to five people hours to kind of curate that and give that [analysis].”
When asked what commercial frontier AI models supported the work, he told DefenseScoop that the team used Palantir’s AI Hivemind product in MSS, which orchestrates, trains and re-trains AI agents.
“We could easily say, ‘Hey, I now want you to take on this persona,’ or ‘I now want you to also add this,’” Samuelson said.
Separately, the embeds also created tools that fused data and analytics on munitions use to help the operators anticipate potential demand and start positioning supplies before the formal requests arrived.
Samuelson said many of the details about other bots and apps that were generated in the operation are classified. He mentioned tools built to address gaps in the Transcom fuel team’s data and analysis capabilities, as well as data streams from aircraft and watercraft downrange about health and maintenance that were changing rapidly in real time.
Because Transcom leads a complex, global logistics network to move, equip and support the joint force, officials knew that when major military operations launch, the command would receive heaps of requests from DOD components about how combat power would be aggregated.
“What we did for OEF — and really had started prior to that — is, ‘I don’t get phone calls, and I’m not going to reply with PowerPoint and emails. I’m going to send you a link that has all of that information.’ So, for [U.S. Central Command], on the Centcom commander’s dashboard, are actual datafeeds that were created at Transcom that are still on our dashboards — but we just open up a link and now they can see the exact same stuff. They don’t have to recreate it. We don’t have to push them information,” Samuelson explained. “It’s live data that’s available to both the Transcom commander and any supported combatant command.”
He said “about half” of the 33 bots and tools Transcom prototyped early into the operation are still “in production” and being used by military personnel today.
“If we want to work at the speed of war, then we need to be able to see that constantly and fine tune our processes and day to day, so we can rapidly scale to war and we are better integrated,” the Transcom CDAO told DefenseScoop.
