ORT/DEF
Systems that hold the mission.
Decision advantage, autonomy, and edge AI for contested, denied, and degraded environments — software and hardware engineered as one capability, delivered as client-owned IP with a human in command.
One system across the battlespace.
How the system connects.
Mission planning, COA comparison, force laydown, and decision audit trails. The operator sees ranked options — the system never decides.
- Mission planning suite
- Decision support engine
- Common operating picture
- Decision audit trails
Fused intelligence picture, threat assessments, ranked recommendations
Commander's intent, collection priorities, rules of engagement
ENGINEERING · DEEP DIVE
The hard problems we solve.
Inference at the edge
Running production ML models on power- and size-constrained edge hardware. We optimize for TensorRT and ONNX, quantize for the target GPU, and maintain accuracy under thermal throttling — because a model that can't run on the fielded device doesn't exist.
- — INT8/FP16 quantization pipelines
- — TensorRT and ONNX Runtime optimization
- — Thermal-aware inference scheduling
- — Model distillation for edge targets
Sync without connectivity
The system keeps operating when SATCOM drops. Local decision authority, conflict resolution on reconnect, and priority-based data queuing — designed for hours or days of disconnected operation, not minutes.
- — CRDTs for conflict-free data merge
- — Priority-based store-and-forward
- — Local decision authority with audit trails
- — Bandwidth-aware sync on reconnect
One codebase, field to cloud
The same application architecture runs on a ruggedized tablet, an edge node, and a cloud instance. Containerized services adapt to available compute, and the UI renders for the bandwidth and display it has.
- — Containerized microservices (K3s to EKS)
- — Adaptive UI for bandwidth and display
- — Shared schema across deployment targets
- — Offline-first data layer with cloud sync
Assurance before the field
Every AI output is evaluated, red-teamed, and traced before deployment. Not a compliance checkbox — a repeatable engineering process that produces evidence a program office can defend.
- — Automated evaluation harnesses
- — Red-team simulation frameworks
- — Model cards with provenance tracking
- — Audit-ready evidence pack generation
Hardware and software on one release train
Firmware, PCB revisions, and application software ship together on a single coordinated release. Hardware-in-the-loop testing catches integration failures before the field does.
- — Unified CI/CD across firmware and app
- — Hardware-in-the-loop test automation
- — Coordinated OTA field updates
- — Version-locked hardware-software manifests
Mission briefs