Recently, our team had the opportunity to work on a project that won the Grand Prize at the Republic of Korea Army AI Competition.
Around the same time, we also participated in the National Security Hackathon in San Francisco, where we explored ontology-mirroring edge devices and Foundry-based streaming workflows. Together, these experiences led to some interesting insights.
One of the challenges we encountered was not AI itself, but the battlefield network.
As drones, ISR platforms, and sensors continue to proliferate, the amount of available data is growing rapidly. However, tactical communications infrastructure remains constrained. In many operational environments, especially where LTE or fixed infrastructure is unavailable, communication relies heavily on RF-based tactical networks.
Under those conditions, transmitting large volumes of video data in real time becomes increasingly difficult.
The question we explored was simple:
What if we transmitted information instead of video?
Using NVIDIA Orin Nano-based Edge AI Vision, we designed a workflow that detects and interprets objects and events directly at the edge, then converts them into semantic information.
Instead of transmitting full-motion video, the system generates observations such as:
-
Five dismounted personnel detected at a specific location
-
Three vehicles moving southbound with estimated speed and position
-
Suspicious activity detected within a designated area
These observations, along with associated metadata, can be transmitted over RF networks with significantly lower bandwidth requirements.
We then used Palantir Ontology to structure and connect these observations, linking entities, events, locations, and timestamps into a decision-support workflow.
One of the most interesting lessons from the project was that the real challenge was not moving data faster—it was transforming observations into information that could support decisions.
Building a realistic testing environment was another challenge.
To simulate operational conditions, we reconstructed portions of the Battle of Avdiivka using publicly available sources, including Telegram, X, news reports, OSINT datasets, and battlefield footage.
This allowed us to build and test a complete pipeline:
Edge AI → RF Transmission → Ontology → Decision Support
Throughout the project, Palantir Ontology served not merely as a data repository, but as a framework for connecting fragmented observations into a common operational picture.
Later, during the National Security Hackathon in San Francisco, we had the opportunity to experiment with ontology-mirroring edge devices and Foundry-based streaming environments.
Seeing data generated at the edge flow into Ontology and then support operational decision-making was particularly compelling. It highlighted the potential of combining Edge AI with ontology-driven systems in constrained environments.
The project ultimately received the Grand Prize at the Republic of Korea Army AI Competition.
For me, however, the most meaningful outcome was not the award itself. It was the opportunity to build and validate an end-to-end pipeline that transforms battlefield observations into decision-ready information.
I’m curious whether others in this community have explored Foundry, AIP, or Ontology in edge environments, contested communications scenarios, or bandwidth-constrained systems.
If there is interest, I’d be happy to share more details on the architecture, data modeling approach, and experimental setup.
We’re also hoping to expand these kinds of experiences within Asia.
This July, Seoul will host D4D (Deploy for Defense), an APAC defense-tech hackathon where builders, operators, engineers, and founders will explore real-world challenges using technologies such as Foundry, AIP, Ontology, and Edge AI.
https://luma.com/2ew4xn7b
Inspired by defense-tech hackathons in San Francisco and Europe, we’re excited to see more opportunities for builders across Asia to experiment, collaborate, and tackle mission-critical problems together.