The platform
Every KBM Air product sits on the same foundation, which is why a risk driver on a dispatcher’s screen and a figure on an executive dashboard mean the same thing. It is also why the honesty rules below are architectural rather than editorial — they are enforced in the code, not in the copy.
Principles
Each of these has an implementation detail attached, because a principle you cannot check is a slogan.
Every fact a KBM Air product surfaces is traceable to the record it came from — an observation, a crew state, a maintenance entry, an advisory. The analyst layer explains risk; it does not invent it.
How it is actually done In Nexus Dispatch and Airline Operations Intelligence the answer path is deterministic: with no language-model credential configured at all, the products still answer, from the records. A model, when present, only phrases what the records already say.
Safety-relevant text and safety-relevant scoring are produced by explicit rules, not by generative models. The same inputs produce the same output, every time, and every phrase traces to a field.
How it is actually done AeroVox renders standard phraseology from a parsed observation through per-language rule engines with no generative step in the text path — which is what makes the multilingual claim defensible to a regulator rather than merely impressive.
No KBM Air product takes a safety-relevant action. Aircraft release, flight-plan filing, broadcast transmission and passenger rebooking all require a qualified human, and the approval is recorded as a human act.
How it is actually done Approval gating is enforced in the engine, not added at the interface: advisory actions append to a decision record naming the person, the time and the reasoning. Higher autonomy tiers exist as designs and stay there until an operator sets the policy.
Decisions are written into a tamper-evident record. A reviewer can verify the chain in the interface — and can watch verification fail on a record edited after the fact.
How it is actually done Records are hash-chained with a keyed signature over the row fields, so the property holds independently of the database underneath. The tamper test is part of the demonstration precisely because a claim you can only assert is not evidence.
Cloud when speed matters, on-premises or inside a controlled environment when control does. The same software, sized to the operator, without ripping out what already runs.
How it is actually done Demonstration surfaces are self-contained with relative paths and no mandatory external service, so they run on a subpath, a dedicated host, or an environment with no outbound access. Secrets live in environment configuration, never in the source.
Maturity is labelled, gaps are named, and no figure appears without the tier of evidence behind it. An overstated claim does not survive a procurement review, so we do not make one.
How it is actually done Every product page carries a “what is not built” list written by the people who built it. The evidence framework is published so a reader can hold each claim to its stated tier.
Architecture
Surfaces, governance and intelligence are built and can be exercised. The data and integration layer is where the real dependency sits, and pretending otherwise would waste your evaluation time and ours.
The screens each role actually works from — dispatcher board, executive rollup, passenger view, broadcast page. Different products, same records underneath.
Human-approval gating, written reasoning on every automated decision, and a tamper-evident chained record of what was decided.
Deterministic scoring, constraint solving and phraseology rendering. Rules first; a model only ever phrases what the rules already produced.
Public aviation weather and published navigation data are live today. Airline system, reservation and operational-feed adapters are written to published schemas and run as labelled simulations until credentials exist.
Deployment
The demonstration surfaces carry no mandatory external service dependency, which is what makes the third option below real rather than a sales answer.
Fastest path to an evaluation. Runs on infrastructure we operate, on synthetic or public data, for as long as the evaluation needs.
Deployed inside the operator’s own infrastructure. No mandatory external service dependency, and secrets stay in the operator’s environment.
For operators whose data cannot leave a boundary. Self-contained deployment with no outbound requirement, intended for inspection before it is trusted.
What none of these modes includes today: single sign-on, role scoping, a data-retention policy and a model-governance programme. Those are pilot-hardening work, and they are named on every product page rather than implied by an architecture diagram.
Security posture
Neither certification programme has started. Both are costed line items driven by a real procurement requirement — which is a different thing from being “in progress”.
Next step
The parts of this platform worth judging are the ones you can try to break: the deterministic answer path and the audit chain. We would rather you tested them than took our word.
Or write directly to air@kbmsolvedit.net.