Explainable AI · decision support

Turn the spectrum into a decision.

AI EW Decision sits on top of OAK Defense's physics-based EW models and applies machine learning to classify emitters, prioritise threats and recommend a course of action — transparently, offline, and grounded in open published literature.

Request a demo Watch the teaser

UNCLASSIFIED // FOR TRAINING USE ONLY — classification-ready, never confers classification

AI EW Advisor

Classifies every emitter, scores threat & priority, and recommends an EW course of action — with a confidence score, the feature that drove the call, and the published source behind each class. No black box.

AI Model Trainer

Generate datasets from the emitter physics and train/evaluate classifiers (accuracy, confusion matrix, ROC) — or benchmark on real public datasets, RadioML and RadChar. Transparent, pure-numpy.

AI EW Assistant

Ask in plain language — "classify the emitters", "highest-priority threat", "what Chinese radars do you know?" — answered from the suite's models, with an open named-system library of 77 systems across 17 countries. Cites its sources.

Cognitive-EW Agent

A closed-loop agent that senses every emitter, learns which technique beats which emitter class, acts and adapts — out-performing a fixed doctrine once it has learned the threat picture.

New · assured extensions

Target recognition & runtime assurance

Two additions that keep the depth deterministic and the AI accountable — imaging you can inspect, and a monitor that grades the model while it runs.

ISAR target recognition

Deterministic inverse-SAR imaging (CLEAN · polar-format · autofocus) forms the target signature, and the existing explainable classifier reads it for automatic target recognition — strong accuracy on held-out test data, with visible confidence and the feature that drove the call. Representational, offline.

Runtime AI-assurance monitor

Watches the model in operation — drift detection, fault root-cause analysis and reliability-corridor re-validation scheduling. Reliability is measured and gated, never asserted, and every decision is written to an evidence-grade, tamper-evident audit aligned to ISO/IEC 27037 and Canada's CGSB 72.34.

Classification-ready

Field it controlled — with your own data and authority

Marking banners, role-based access, an audit log and bring-your-own-data let you operate it in a controlled environment. The software displays the handling level you assert — it never confers classification.

AI EW Decision — classification banner & sign-in

UNCLASSIFIED // FOR TRAINING USE ONLY

CUI

CONFIDENTIAL

SECRET // NOFORN

TOP SECRET

TOP SECRET // SCI

Operator-asserted handling banners. Everything shipped is UNCLASSIFIED // FOR TRAINING — no controlled data, no operational technique library, no exploit content.

Why it wins evaluations

Honest by design

Explainable

Visible confidence and per-feature contribution for every classification — auditable, not a black box.

Cited & grounded

Taxonomy & features trace to RadChar (ICASSP 2023), RadioML 2018, MIT OpenCourseWare, Skolnik / POMR, FOI. The assistant shows its references.

Real-data proof

Benchmarks the same models on public RadioML / RadChar datasets — in transparent, pure-numpy code, fully offline.

13-class taxonomy Named systems · 17 countriesRBAC + audit Bring-your-own dataOffline single-exe

Put it on your threat picture

See the Advisor classify and recommend, the Assistant answer in plain language, and the agent learn — on a scenario you care about.

Request a demo