AI that works at sea, not just in the demo.
Most maritime AI assumes a fast connection, clean data and someone with time to babysit it. Vessels have none of those. We build AI for the conditions you actually operate in.
Everyone is selling you AI. Almost none of it survives a voyage.
The pitches sound the same. A dashboard, a chatbot, a promise of predictive maintenance. Then it meets a real vessel: a satellite link with 400 ms latency and a monthly cap, a maintenance history typed by fourteen different chief engineers over twenty years, and a crew who already have a job.
The result is a pilot that impresses in the boardroom and quietly dies at sea.
We start from the other end. What is actually costing you money, what data do you actually have, and what will still work when the vessel is three days from anywhere with no connection.
Four things AI is genuinely good at in shipping.
Not everything. Four things. We would rather do these properly than sell you a platform.
Reading your documents
Your fleet drowns in paper. Certificates, class documents, PSC reports, invoices, charter parties, survey reports. Most of it is scanned, unsearchable and answered by someone remembering where they filed it. Modern language models are extremely good at this.
- Certificates read automatically, with expiry dates tracked before they bite you
- PSC and vetting reports parsed into findings you can search across the fleet
- Invoices matched against orders without someone doing it by hand
- Ask a question in plain language and get an answer with the source document attached
Typical payback: the fastest of everything on this page. Usually weeks, not quarters.
Predicting failures before they happen
If you are collecting sensor data, you are sitting on a signal. Bearings, pumps, compressors and separators rarely fail without warning. They drift first. People miss the drift. Models do not.
- Condition monitoring on the equipment that actually hurts when it stops
- Early warnings with enough lead time to order the part and plan the work
- Maintenance moved from calendar-based to condition-based, where it makes sense
- Honest confidence levels: the model tells you when it does not know
Requires sensor data. If you are not collecting it yet, we will tell you that first, and help you start.
Scoring risk before the inspector does
Port State Control detentions are predictable more often than the industry admits. So is vetting failure. The patterns are in your own history and in public inspection data.
- A risk score per vessel, updated as conditions change
- The specific findings most likely to be raised, so you can fix them first
- Prioritised pre-arrival checklists based on the port you are actually going to
Answering questions across your fleet data
Your operational knowledge is spread across a fleet management system, a maintenance history, a document store, a dozen spreadsheets and several people's heads. We build assistants that sit across that data and answer in plain language, grounded in your records, with citations.
- “Which vessels are overdue on lifeboat drills?”
- “What did we do last time this separator failed?”
- “Show me every deficiency raised at Rotterdam in the last two years”
The model that needs the internet is the model that fails at sea.
This is the part most AI consultancies cannot do.
Almost every AI product on the market sends your data to a cloud somewhere and waits for an answer. That is fine ashore. On a vessel it means your AI stops working exactly when the connection does, and it means your maintenance history, crew data and commercial documents leave the ship.
We build AI that runs on board. The model lives on vessel hardware. It works with the satellite link down. Your data never leaves the vessel unless you decide it should.
We designed our own maritime software offline-first. We build AI the same way.
- It works offline. Mid-ocean, in a dead zone, during an outage.
- It costs nothing to run. No per-query cloud bill, no bandwidth spent on inference.
- Your data stays yours. Commercially sensitive information stays behind your own firewall.
- It fits your cyber posture. Fewer external dependencies is a shorter attack surface, which matters as IACS UR E26 and E27 land.
We start small, on purpose.
AI projects fail when they start big. We would rather prove something works on one vessel than promise something across thirty.
Discovery
2 weeksWe look at your data, your systems and your costs, and tell you honestly where AI pays and where it does not.
You get: A written assessment with ranked opportunities, effort and expected return.
Fixed feePilot
6-8 weeksWe build one thing, on one vessel or one workflow, and measure whether it worked.
You get: A working system and a straight answer on whether to continue.
Fixed feeBuild
Scoped per projectWe build it properly, integrate it, and put it into daily use.
You get: Production software your team can actually use.
QuotedSupport
OngoingWe monitor, retrain and improve it as your fleet changes.
You get: A model that stays accurate instead of quietly rotting.
RetainerNo lock-in. After Discovery you can take the assessment and build it yourself, or with someone else. It is your document. We would rather be chosen than trapped into.
Some honesty about the limits.
The fastest way to trust a vendor is to hear them say no.
We will not touch safety-critical navigation.
Collision avoidance, autonomous control, anything where a wrong answer puts people in the water. That is not where a consultancy should be, and anyone who tells you otherwise should worry you.
We will not sell you AI you do not need.
If Discovery shows the answer is a better process and a cleaner database, we will say so and hand you the invoice for two weeks instead of two years.
We will not train models on your data for someone else's benefit.
Your operational data is a commercial asset. It stays yours, and it does not end up in anyone's training set.
We will not promise accuracy we cannot measure.
Every model we ship comes with its error rate written down. If we do not know how often it is wrong, we do not ship it.
We build maritime AI because we build maritime software.
This is not a consultancy that discovered shipping when AI became fashionable. Our founder has spent 22 years in maritime IT: vessel networks, fleet systems, and the unglamorous reality of making technology survive on a ship.
We build and run our own maritime products, with AI features in them. That means when we advise you, we are describing things we have already had to make work, not things we have read about.
We understand the regulation too: ISM, PSC, vetting, CII, FuelEU, IACS UR E26/E27. And we are offline-first by default: on-board inference, not cloud dependency.
Find out where AI actually pays.
A short call, no cost. Tell us what your fleet looks like and what is expensive. We will tell you honestly whether AI helps, and where we would start.
Questions we get asked
We do not have good data. Is that a problem?
It is the normal situation, and it is the first thing we look at. Some AI needs clean structured data. Document intelligence does not; messy scanned paper is exactly what it is built for. We will tell you which category you are in.
Will this work with our existing fleet management system?
Usually. We integrate with what you have rather than asking you to replace it. If your system has no API at all, we will say so early.
Does our data leave the vessel?
Only if you want it to. We build on-board models specifically so it does not have to.
How much does it cost?
Discovery is a fixed fee, and it is deliberately small. Everything after that is quoted once we both know what we are building. We do not quote AI projects blind, and you should be suspicious of anyone who does.
Are you selling us a product or a service?
Both are available, and we will be clear about which one we are recommending and why.
Can you help us with EU and IMO reporting obligations?
Yes. CII, FuelEU and MRV reporting are largely a data problem, and data problems are what we are for.