NEPTUNE DSS SILICON MOUNTAIN

NEPTUNE DSS

The system doesn't give more data. It gives better decisions.

Single-operator command and decision support for unmanned maritime operations: one operator, six vehicles across air, surface and subsurface, protecting ships and seabed infrastructure. NEPTUNE is the decision and accountability layer that sits on top of the combat system. Deterministic, offline-capable, no LLM in the decision loop. TRL 4-5 working prototype.

Already have an access link? Open it in this browser.

Five decision functions, declared

Not generic AI. Each recommendation comes from a named objective function with real constraints and an exact or tractable solve, and the explanation is generated from the computation itself. The engine runs deterministically, offline, with no LLM in the decision loop.

D1Re-tasking
A vehicle fails, the fleet re-matches to tasks. An assignment problem solved exactly with the Hungarian algorithm, maximising retained mission value under capability, reach and energy constraints.
D2Re-planning
The mission plan is rebuilt around what survives: which tasks are kept, which are dropped, the completion confidence of each, and the trade-off stated in plain terms.
D3Triage
Alerts are scored relevance × criticality × urgency and presented one at a time. Multiplicative, so no single axis hijacks the queue.
D4Uncertainty
Every track carries a full position covariance that grows honestly when contact is lost. Subsurface detection range is the sonar equation, computed in decibels, not a guess.
D5Trade-offs
Candidate courses of action reduce to the non-dominated set over time, energy, coverage and risk: a Pareto front the operator explores before committing.

Honest uncertainty. Signed decisions.

When two sources disagree about the same contact, NEPTUNE does not average them into one smooth label. Identity is fused with Dempster-Shafer belief theory: every source keeps an explicit ignorance mass, and every combination yields a conflict figure alongside the fused belief. The disagreement reaches the operator as a record: both claims, which source is trusted, and why. A self-reported identity collapses the moment a measurement contradicts it.

Every committed decision becomes a sha256 hash-chained decision-of-record. It binds the evidence the operator saw and the operator's signed identity: who decided, not just what. Edit any past entry and verification breaks. The chain exports whole for third-party verification.

CONFLICTK 0.09
AISself-reports "fishing"
RADARmeasures 26 kn
TRUSTradar. AIS weight collapsed on contradiction
FUSEDfast_inflatable, bel 0.47, pl 0.81

Surfaced, never averaged away.

DECISION-OF-RECORDseq 0147
PREVsha256 3aa1f0…9c42
HASHsha256 e7b104…1d8f
OPERATORop-01, identity signed into the hash
EVIDENCEgeometry, rationale, alternatives rejected

Tamper with any past entry: the chain fails verification.

Live data, anywhere the map goes

The console ingests live open data: real aircraft, ship traffic, seismic events, and the marine weather that drives the acoustic model. The feeds follow the viewport worldwide: pan the map to any coast on Earth and every live axis re-queries that region.

ADS-B

Live aircraft, real callsigns and altitudes.

AIS

Ship traffic, with a synthetic fallback that says so.

SEISMIC

Live events, raising cable-integrity alerts.

MARINE WX

Sea state feeding the acoustic model.

Live feeds pass the same validation gate as simulation inputs. Every feed carries a health state, a cache and a fallback: kill one at runtime and the console keeps working on cached, extrapolated data, and says exactly what happened.

How a demo works

  1. Write to us. You receive a private access link, scoped to you.
  2. The link opens an isolated live session in your browser, running the full console.
  3. Nothing to install. When your session ends, it is gone.

Request a live demo