AWS showcases generative AI for real-time military logistics

In this demo, we have a PACOM logistics assistant that takes geospatial data and then overlays our telemetry and our information for bases and ships to let the battle commanders ask questions in an evolving operations

Hi, my name is Cheng Ho, and I am a Senior Solution Architect with Amazon Web Services

Today, I'm presenting a solution created with Genitive AI that uses geospatial information

together with retrieval augmented generation. We took data from planes, telemetry, how far it can fly, the weight, along with the information for bases

Basically, if it could support a certain plane, the distance between, and so that we can support operations

As battle commanders, they can ask additional questions of the AI system, and they get answers in real time

This RAG system integrates with existing GIS and legacy systems by using a default or universal standard for coordinating information

So we can take any information, top graphical maps and overlay information that the genitive

AI brings for us So it depends on if you know a certain mountain it has been by a unit we can use the height and distance to estimate how far a tank can go on top of it

Obviously, the higher it is, the less the tank can go, and then use that information to support the Battle Command

So in areas of limited connectivity and also resources, what we use is a reduced model

So instead of using the full-blown, like a heavy model, like 120 terabyte open AI model, we can use like a 20 terabyte model so that it's much more lightweight

You get about 85% of the quality of answers, right, with maybe about 10% of the size

System performance and data accuracy are monitored through a CloudWatch dashboard. So all the integration units, points that go into the system will be reported to CloudWatch

weather and response time to be reported. We can see any latency before it develops

As questions are asked, we can see if the inference engine is slowing down and then account for it by adding more resources if necessary

So we've had battle commanders take a look at the system and attempt to use the system

whether it be text-based or whether it be something that's more topographical map-based

They said that the solution that is proposed or that is shown the actions that are presented and information that shown humans can come up with it themselves Normally it takes 15 maybe 20 minutes for someone to figure that out and do the math for it Whereas the logistics agents is what we calling it can come up with information in seconds

In this demo, we have a Paycom logistics assistant that takes geospatial data

and then overlays our telemetry and our information for bases and ships

to let the battle commanders ask questions in an evolving operations. So let's say, for example, I'm looking here and I have a ship that's the C-130J

And I want to know what bases can the C-130J land on

So I'm looking at the system here. So I'm going to go ahead and type in the question. So the question I have here is, what bases in the PACOM region can a C-130J take off and land from

So I'm going to execute it. And it should tell me the list of bases where I can actually land a C-130J because it is a different type of plane

does tuning support, size, and whatnot determine where it can land. So there we have the various bases that the plane can land and take off from

So now I'm going to ask about the range of where it can go

So here I'm asking the system to draw me a circle or draw me the range of how far a C-130J can

with the default amount of view, where it can, taking off from Hickam Air Force Base, how far it can go

As you can see here the system itself has outlined the range here and it not a full circle because it accounting for the curvature of the earth right so that a little bit more of a pick shape but things don stay static right we can add or

move things to a C-130J so let's ask a question like what if I have an external few pack on top of

it because sometimes we want extended range so I'm going to ask about what if we put in a 17,000

pound few pack on top of the cone so as you see here the the green circle was our original range

Now the pink circle is the range of the 17,000-pound fuel pack on top of it

So as you can see, using the system, which we've created very rapidly, within a day or so

to rapidly prototype what AI can do, we've enabled instant questions from battle commanders

asking about the operations, and also changing the parameters as they go along. So this is just one example of how Gen AI can assist in the operations in the PICOM environment

The system is extensible. We can add in additional planes, ships. Not only that, we can add in military vehicles and then also enemy units also

This is just one example of using AWS and our generative AI abilities to rapidly prototype and create applications to support operations in the Indopaycom environment or, for that matter, any environment