Generic AI chatbots are impressive, but in a ship's engine room ‘plausible’ is not good enough. An answer about a purifier or a governor must come from the right manual, for the right model, on the right vessel — and the engineer must be able to check it.
What RAG actually is
Retrieval-augmented generation splits the job in two. First, a retrieval step finds the most relevant passages from your own approved documents — OEM manuals, drawings, SOPs, service reports, maintenance history. Second, a language model writes an answer using only those passages, and cites them.
Why scoping matters at sea
A fleet may have dozens of generator models. Scoping retrieval by company, vessel and equipment means a question about ‘Generator 2’ retrieves that generator's manual and history — not a similar-sounding passage from a different maker.
Scanned manuals are the hard part
Much marine documentation exists only as scanned PDFs. Page-by-page OCR, combined with native text extraction, turns those scans into searchable knowledge — and keeping page references means every answer can link back to the exact page.
A worked example
- Ask: ‘Why is Generator 2 exhaust temperature increasing?’
- Investigate: retrieve current and historical data for that generator.
- Learn: find similar incidents and the corrective actions that worked.
- Reference: open the relevant OEM troubleshooting pages.
- Check: review maintenance history and work performed.
- Respond: present a source-grounded troubleshooting view.
- Act: create or follow up the appropriate PMS job.
Privacy and model choice
Technical documentation is commercially sensitive. A marine RAG system should keep your documents in your own environment, enforce role-based access to them, and let you choose the answering model — including private or self-hosted models.
MARKS, the NauticalFlows marine knowledge system, is built on exactly these principles.