WxCI is building explainable, multi-resolution-quantized AI at the edge — turning raw sensor streams into auditable decisions on microcontroller-class hardware, with no cloud dependency. Host-validated today; hardware bring-up and <100 ms latency measurement are our current engineering milestones.
The Boreal Technologies stack is a neuro-symbolic decision engine: neural perception handles noisy real-world signals; a symbolic policy layer makes the call — every output explainable, versioned, and remotely governable.
Acoustic · seismic · thermal · environmental telemetry (driver integration in progress)
Multi-resolution transform features — tuned for transients
INT8 quantized networks, <20 KB — on-MCU inference is the active build item
Symbolic rule engine with versioned policy ontologies
Decision + confidence + rule trace → your C2, SCADA, or alerting plane
Explainable by construction. Every detection carries a rule trace and a versioned-policy decision record — the artifact regulators, auditors, and commanders actually require.
Logic updates, not retraining cycles. A verified over-the-air channel for ontology and model updates (CRC32 + SHA-256 today; cryptographic signatures and radio transport are the next integration items).
Runs without the cloud. Policy and decisions execute on-device by design; degraded-comms behaviour is a Stage-1 test criterion.
The same multi-resolution transform mathematics targets the models themselves: weights stored in a multi-resolution representation (.bwave format, implemented and verified with a 93-check round-trip suite), reconstructed layer-by-layer in integer-only arithmetic. Large-model compression payoff (2–4× targets) is our Stage-2 research objective.
Multi-resolution weight format (.bwave) with scale-wise precision allocation.
Integer-only inverse transform runtime — no FPU, shift-and-add only (Stage-2 design).
Sovereign, auditable sensor intelligence for contested environments: our detection/classification core is host-validated, and mesh, C2 integration, and GNSS-denied behaviour are the active engineering roadmap — currently proposed through DND IDEaS programs.
Edge condition monitoring for rotating equipment, conveyors, and remote sites: multi-resolution feature extraction detects faults and transients early; policy ontologies enforce site-specific alarm logic; decisions arrive with evidence for safety case compliance. No connectivity required.
The WxCI pedigree. Weather, climate, and hydrological signals — water levels, air quality, thermal and seismic events — classified on-device by design, with explainable outputs — reference build validated in host testing; pilot deployments under discussion.
Small enough to hand to a field crew. Honest enough to sign your name to.
Weather Climate Insight Inc. is a Canadian-incorporated company. Our decision intelligence was born in environmental sensing — where forecasts, risk, and trust in a signal are the business — and hardened for the environments where connection to the cloud cannot be assumed: defence operations, remote mine sites, and territory-scale monitoring networks.
All IP vests in WxCI. Canadian-controlled, open-standards, no walled gardens.
• DND IDEaS minefield reconnaissance submission (Oct 2026, $850K ask)
• DND CUAS 2027 Urban Sandbox application (Dec 2026; live demonstration July 2027, Ottawa)
• IDEaS CFP6-CH14 sensor-fusion middleware (Mar 2027)
Builds clean today on the official Portenta H7 toolchain with a real-time OS and a fixed-point DSP library vendored, covered by 8 host test files (all passing, run log committed) — the on-device bench run is the next milestone. Evidence register: docs/AUDIT_CLAIMS_GAPS.md.
Whether you operate sensors in a contested theatre, a mine site, or a watershed — we will show you the stack running on real hardware, with real decisions, in under an hour.
Explainable edge intelligence — sovereign by design.
Pilot programsWe run 90-day field pilots: one site, your sensors, a working control plane. Defence, mining, and environmental operators welcome.