N51 and IDEaS Challenge 13: How the KANATA stack meets Canada's multi-modal AI requirements
DND/CAF Challenge 13 defines exactly what Canada's armed forces need from an AI system. This is our direct technical response, product by product, requirement by requirement.
What Challenge 13 actually asks for.
In June 2025, the Department of National Defence published IDEaS Challenge 13: Multi-Modal AI for Advanced Situational Decisions. The challenge asks for a system that can fuse heterogeneous sensor data from at least two modalities, surface explainable AI-driven decisions in real time, and do all of this within the size, weight, and power constraints of tactical edge environments.
We read it and recognised every requirement. Not because we adapted our platform to meet them because we had already built against these problems from the first line of code. KANATA, INUKSHUK, KITCHI, and MISTIG are the direct answer to Challenge 13. This piece walks through how.
The essential outcomes, addressed.
Challenge 13 identifies three essential outcomes. A qualifying system must fuse at least two heterogeneous data streams in real time. It must surface explainable outputs that operators can interrogate and trust. And it must operate within SWaP-constrained environments — meaning tactical edge hardware, not cloud infrastructure.
INUKSHUK addresses the first requirement directly. The fusion engine ingests RF intercepts, EO/IR camera feeds, satellite imagery, ground sensor telemetry, acoustic data, and open-source intelligence simultaneously — seven modalities fused into a single spatiotemporally aligned common operating picture in under 100 milliseconds. The pipeline runs on edge hardware with no cloud dependency.
KITCHI addresses the second. Every course-of-action recommendation KITCHI surfaces includes the reasoning that produced it — which sensor inputs triggered the assessment, what confidence level the model assigns, and what the uncertainty range is. A commander looking at a KITCHI output sees not just what the AI recommends, but why. That is the explainability requirement met at the architectural level, not the interface level.
KANATA addresses the third. The full pipeline — ingestion, fusion, reasoning, output — runs on NVIDIA Jetson Orin-class hardware. No satellite uplink required for core operations. Full mission capability in degraded and denied connectivity environments. Sovereign infrastructure end to end.
“A system that makes the right recommendation for the wrong reasons is dangerous. KITCHI shows its reasoning on every output — not because we were asked to, but because a commander who understands why is a commander who can override when it matters..”
Beyond the essential outcomes, Challenge 13 identifies a set of desired capabilities that a qualifying system should demonstrate. These include policy-aware data fusion with classification-level provenance tracking, entity resolution and persistent knowledge graphs across sensor modalities, and a scalable pipeline that can onboard new sensor types without re-architecture.
MISTIG handles classification-level provenance. Every data element flowing through the KANATA pipeline carries metadata tracking its source, its classification level, and its handling history. When a SECRET-level satellite feed fuses with an UNCLASSIFIED ground sensor, MISTIG automatically elevates the output classification and tags it accordingly. No manual relabelling. No human error in the classification chain.
INUKSHUK handles entity resolution. Entities detected across different sensor types — a vessel appearing on radar, then on AIS, then on EO/IR — are correlated and maintained as persistent tracks rather than re-identified on each new detection. The knowledge graph grows with every sensor pass.
KANATA's modular ingestion layer handles scalability. New sensor modalities are onboarded via a standardised API. The platform does not require re-architecture to accommodate a new data source. When NANOOK, our aerial surveillance platform, comes online — it plugs into KANATA natively.
What we're asking evaluators to review.
N51 is an MVP-stage company. The KANATA stack is built, tested, and operational. We are actively seeking DND/CAF evaluation engagement ahead of the IDEaS Challenge 13 deadline.
We offer two levels of engagement. An unclassified demonstration covers the full KANATA architecture, live INUKSHUK fusion across multiple sensor inputs, KITCHI decision-support outputs with visible XAI reasoning, and MISTIG's security layer overview. A classified technical briefing covers the full stack at the appropriate classification level, including sensor integration specifications, edge hardware configurations, and MISTIG's cryptographic architecture.
If you are a DND/CAF evaluator or IDEaS Challenge 13 reviewer, contact us directly at inquiry@n51.ai. We respond to every inquiry personally.
Sovereign by design.
Canada has the talent, the mandate, and the terrain to build sovereign intelligence infrastructure. Challenge 13 defines the requirement. N51 builds the answer. We look forward to demonstrating it.
