AquaPilot — AI Marine Co-Pilot

Product Overview

Commercial-in-confidence. Access is granted per viewer and is subject to a confidentiality agreement.

Access is logged. Your name, organisation, email, electronic signature, acceptance time, agreement version and network information are recorded for security and record-keeping.
Smart-Tech Enterprise Solutions
v1.0

Confidentiality & Non-Disclosure Agreement

By accessing this presentation, you agree that all information, materials, designs, concepts, technology, business information, documents, images and other content made available to you (Confidential Information) are confidential and belong to Smart-Tech Enterprise Solutions Pty Ltd.

You agree to

  • keep all Confidential Information strictly confidential;
  • use it only for the purpose of reviewing or discussing this presentation with us;
  • not share, publish, forward, reproduce, photograph, screenshot, screen-record or distribute the presentation without our written permission;
  • not upload or submit Confidential Information to third-party or public AI platforms without our written approval;
  • not use the Confidential Information to copy, reproduce or develop a competing product, service or solution; and
  • take reasonable steps to prevent unauthorised access or disclosure.

Exclusions

These obligations do not apply to information that is already publicly available through no breach of this Agreement, or that you can demonstrate was lawfully known to you independently.

Intellectual property

All intellectual property and ownership rights remain with Smart-Tech Enterprise Solutions Pty Ltd. Access to this presentation does not grant you any licence or ownership rights.

Term

Your confidentiality obligations continue for 5 years from the date you accept this Agreement, or for as long as the information remains a protected trade secret where applicable.

Acceptance

By entering your name, providing your electronic signature and clicking “Agree & Continue”, you confirm that you have read, understood and agree to be bound by this Confidentiality Agreement.

Record keeping

For record-keeping and security purposes, we may securely record your name, electronic signature, date and time of acceptance, agreement version, IP address and related access information.

Governing law

This Agreement is governed by the laws of Victoria, Australia.

— END OF AGREEMENT —
Scroll to the end to continue
Sign here

Agreement accepted

Your acceptance has been recorded.

01 — The problem

The data is already there.
Nobody can reach it.

A modern recreational boat carries a live network of depth, engine temperature, oil pressure, fuel, RPM, position and dozens of alarm channels. Almost none of it reaches the person actually steering.

Data out of reach

It's behind a menu

Reading any of it means taking your eyes off the water and navigating a proprietary screen, in different units, in different menus.

No signal

Connectivity ends at the ramp

Cloud assistants and app dashboards are useless where boating actually happens. Most consumer AI degrades to nothing without a signal.

Knowledge in paper

The answer is in a manual

Oil specification, vessel draft, trim operation, local hazards, tide behaviour — spread across service manuals and local knowledge, none of it searchable at the helm.

Cognitive load

One person, four jobs

A solo or family skipper is navigator, engineer, lookout and safety officer at once. Routine checks get skipped because there is no second pair of hands.

And one problem that every marine-AI concept underestimates.

Language models confabulate. An assistant that invents a tide time, a depth reassurance or a battery state is not merely unhelpful — it is more dangerous than no assistant at all. AquaPilot was built around that problem rather than in spite of it.

03 — What AquaPilot is

Two computers. Two jobs.
One boundary.

A deterministic safety computer decides — with no AI of any kind on the wired alarm path. An AI co-pilot advises, on a box that is explicitly allowed to be switched off. Nothing that must be heard depends on the AI.

System boundary · live data flow NMEA 2000 17 DEVICESENGINE · GPSDEPTH · TANKS BRAIN 1 Safety Kernel DETERMINISTICNO AI · EVER SIGNAL K HUBDECIDES SAFETY PRIORITY-1 ALARM WIRED · <2 SECONDSNEVER BLUETOOTH · NEVER AI READ-ONLY BRAIN 2 · “SKIPPER” Advisory co-pilot VOICE · KNOWLEDGE · CONTEXTMAY BE SWITCHED OFF AI NEVER ACTUATES
Brain 1

Decides

The deterministic layer. Owns the vessel data hub, the safety kernel and the monitors. Bridges the NMEA 2000 network. Contains no AI of any kind.

Brain 2

Advises

“Skipper” — the conversational co-pilot. Voice, local knowledge, tides, manuals, live readings. Runs entirely offline. Never touches a safety decision.

Shore tier

Accelerates

An optional workshop inference tier answers heavier questions faster when the boat is in range. Never required — the boat stands alone by design.

04 — How it stays honest

Most questions never
reach the model at all.

Every question is tested against a chain of deterministic handlers, cheapest and most certain first. The first one that matches answers from code, and the language model is bypassed entirely. Only genuinely open questions reach it — and even then it phrases facts it was handed, not facts it invented.

Routing · “what's the tide doing?” EMERGENCY PROCEDURES COMMANDS & CHECKLISTS LIVE VESSEL DATA WEATHER TIDES ◄ MATCHED TRIP ADVISOR COHERENCE GATE LANGUAGE MODEL — NOT REACHED SUB-SECOND · NO MODEL · NO INVENTION

What that buys you

Determinism where it matters, and speed as a by-product. A guardrailed answer returns in under a second. It cannot hallucinate, because there is no model in the path to hallucinate with.

The system reports hazards. It never reports their absence as safety.

No sensor set can support a claim of safety. Skipper says “Mercury reports no engine alarms” — never “the engine is fine”. Every capability with a blind spot states that blind spot each time it speaks.

Detailed routing logic, handler design and the test corpus are available under NDA.

05 — What is built today

This is not a concept deck.

AquaPilot is installed and operating on its proving vessel. The full conversational loop has been validated live and offline aboard. Below is the honest status — including what is not built.

CapabilityStatusNote
Conversational co-pilot — wake word, speech, knowledge, guardrailsBuiltValidated aboard, fully offline
NMEA 2000 bridge — 17 devices, twin plotters, engine gatewayBuiltInstalled aboard
Safety Kernel — anchor drag, depth, bilge, battery monitorsBuiltLogic complete and tested; validation against live network data pending
First-of-day briefing, interactive checklists, catch marks, self-checkBuiltLogged and archived
Engine, fuel and battery readouts across 26 Mercury alarm channelsBuiltWith staleness discipline on every reading
Trip recorder, maintenance register, barometric historyBuiltArchived to private storage
Route handoff to the chartplotterTransport provenImport validated; the plotter remains the navigational authority
Remote inference tier with automatic local fallbackBuiltFallback demonstrated end-to-end
Wired Priority-1 alarm actuationNot builtAwaiting hardware fit — the single step from safety monitor to safety system
Emergency workflows (man overboard, medical)DraftedVerbatim owner-approved procedures, gated behind full validation
Web dashboard suite (helm iPad and shore)DesignedSix-phase build plan
We publish our own worst behaviour.

A garbled transcript once reached the model and it invented navigational advice. That is logged as the highest-priority open item rather than quietly accepted. A system that records its own failures is one you can trust the rest of. That discipline is the product as much as the software is.

06 — Local-first by architecture

Every rung is a working system.
Safety is identical on all of them.

The kernel never consults a model, so capability degrades gracefully while safety does not degrade at all. This is the property that makes AquaPilot deployable where Australians actually boat.

Safety computer onlyDeterministic kernel, monitors, anchor watch, Priority-1 alertingFull
+ co-pilot, offlineFull voice assistant: knowledge, tides, briefing, proactive nudges — all localFull
+ shore linkFaster answers with richer retrieved contextFull
+ internetWeather cache refreshFull
Data sovereignty

Nothing leaves the owner's network

Vessel data, voice, logs and position stay on hardware the owner physically possesses. No account, no telemetry, no third-party service in the safety path. In a category where “smart” usually means “sends your location to a vendor”, this is a substantive differentiator — and increasingly, a purchase criterion.

Cost structure

Commodity hardware, no cloud bill

The system runs on commodity computing hardware and open-source components rather than licensed marine software. There is no recurring inference cost per vessel in any safety function — a materially different unit economic to any cloud-dependent alternative.

08 — How it reaches other boats

Built for one boat properly,
before any claim of generality.

AquaPilot was deliberately proven on a single vessel to a real standard rather than demonstrated broadly and finished nowhere. That said, the pattern it establishes is portable — and the portability was designed in, not discovered afterwards.

Portability

The architecture is vessel-agnostic

It reads the NMEA 2000 standard that virtually every modern recreational boat already speaks. Adapting it to a new vessel is a configuration exercise, a knowledge-base rebuild from that boat's own manuals, and a curated local destination library — not a re-engineering project.

Fitment

Installed by people already opening the dash

The natural channel is marine electronics installers and repower workshops. The system taps an existing network through a standard gateway; it does not require the boat to be rewired around it.

Running cost

No cloud bill, ever

Commodity computing hardware and open-source components rather than licensed marine software. No per-vessel inference cost, no account, and no subscription attached to any safety function — by architecture, not by pricing policy.

Who it suits

Owner-operators first

Solo and family skippers on instrumented recreational vessels, where one person carries the navigator, engineer, lookout and safety-officer load at once.

Beyond recreation

Fleet, charter and hire-and-drive

Operators handing vessels to people who don't know them, where an assistant that reads the boat aloud and logs every trip answers a duty-of-care need rather than a convenience one.

Beyond boating

The pattern generalises

“Deterministic code owns decisions; AI owns language” applies anywhere an assistant is useful but must never be load-bearing — remote industrial sites, agriculture, aged care, emergency services.

A design whose full capability survives with no connectivity is directly relevant to the hardest parts of Australia to serve.

Not just the water. Anywhere the signal ends before the work does.

09 — Roadmap

Three horizons.
Each one earns the next.

Nothing ships until the layer beneath it is validated. That sequencing rule is the reason the product is trustworthy, and it is not negotiable for commercial reasons.

Horizon 1 · nowThe informed skipper

Complete the safety system and productise it

Fit the wired Priority-1 alarm path and close the input-quality guard. Verify depth datum and draft to lift the grounding interlock. Ship the helm and shore dashboards. Package the install so a second vessel is a configuration exercise rather than a rebuild.

Wired alarm fitSecond vesseliPad dashboardInstall kit
Horizon 2 · nextThe augmented helm

Sensing, heads-up display and fleet

Bilge and fire detection wired to kernel inputs with named-zone announcements. AIS traffic with closest-point-of-approach — stating its blind spots every time. Fuel endurance and return margin. A heads-up display carrying continuous numeric readouts and hands-free procedures, with blank-on-stale behaviour as a hard requirement. Multi-vessel management for fleet and charter operators.

Zone sensingAIS & CPAFuel enduranceAR heads-up displayFleet console
Horizon 3 · the second actAssisted handling

The boat starts to help you move it

Assisted docking, station-keeping and — under strict conditions — summoned movement within a marina. This is a different product class with a different risk profile and a different regulatory pathway. It is set out in full below, including what has to be true before a single line of control code is written.

Assisted dockingStation keepingSummonCertification

11 — Why it is hard to copy

What a competitor
cannot clone in a quarter.

Architecture

The safety boundary itself

A written, tested, version-controlled separation between what decides and what speaks — arrived at by surviving real failures, not by design review. Copying the diagram is easy; earning the judgement inside it is not.

Discipline

The decision record

Seven governing records, including rejected proposals and the founder's own refused feature request. This is the artefact that lets a regulator, an underwriter or a partner audit the system — and almost nobody in this category has one.

Corpus

Vessel and local knowledge

A structured, tiered knowledge base combining vessel documentation, local hydrographic knowledge and operational scenarios. Each new cruising ground is an asset that compounds.

Integration

The last mile is physical

Bridging a real vessel network, an engine gateway and a helm audio path is unglamorous integration work that software teams consistently underestimate. It is already done and documented.

Position

Local-first is a wedge, not a constraint

Incumbent marine electronics vendors are structurally committed to cloud accounts and subscriptions. Building genuinely offline is against their business model — which is precisely why the position is available.

Honest note

The trade-off we accept

Local-first means no fleet-wide telemetry by default. We trade the aggregated-data advantage for trust and sovereignty, and recover what we need through opt-in, owner-controlled sharing. It is a deliberate choice with a real cost, and we would make it again.

12 — What is still open

The list we keep,
and don't quietly shorten.

Every serious system has an open list. Ours is written down, prioritised, and published in the same document as the achievements — because a project that only records its wins is one you cannot calibrate.

Open itemPosition
Wired Priority-1 alarm actuationThe kernel decides and reports, but the wired alarm it would drive is not yet fitted. This is the single step that turns a safety monitor into a safety system, and it is the top of the list.
Input-quality guardA garbled transcript once reached the model and it invented navigational advice — the one unsafe behaviour ever observed. Closing that gate so no unclear speech can reach the model is the second priority.
Depth datum and draft verificationGrounding and anchor-scope logic remain interlocked until the transducer offset and vessel draft are confirmed by wet test. Interlocked, rather than shipped on an assumption.
Kernel validation against live network dataMonitor logic is complete and covered by its own tests. Validation against the running vessel network, under way, is outstanding.
Second-vessel portabilityPortability is designed in and argued for, but not yet demonstrated. Until a second boat is running it remains a claim rather than a result — and it is described that way.
Commercial data licensingWeather and tide data are used under personal-use terms today. Any commercial deployment requires a registered commercial agreement with the relevant agency. Identified and sequenced, not discovered late.
AIS coverage honestyMost vessels in these waters carry no AIS. Every AIS answer must state that limit each time it speaks. A quiet screen must never be heard as “nothing out there”.
Emergency workflowsVerbatim, owner-approved procedures are drafted and frozen, but deliberately gated behind sustained real-world validation of everything beneath them. Releasing them early would defeat their purpose.

13 — Who is building it

Smart-Tech Enterprise
Solutions Pty Ltd

Satheesh Agatheeswaran — founder and principal engineer. Designed and built the complete system: vessel network integration, deterministic safety kernel, offline voice stack, knowledge layer and the governing decision framework.

Founder background, prior ventures and relevant credentials to confirm.

The proving vessel is not a demo rig.

AquaPilot was built for one real boat, used by the person who built it, in the water it was designed for. Every design rule in this document came from something that went wrong on the water first.