Our world seems to change by the day; a new industry has arisen during our build of Vanguard. The advent of AI as an evolution of traditional search engine use. We have a lot of data gathered during our build, what, where, when, and how others did it. So it was time to play a little, and this is how we used an AI Chatbot to help.
A specialized Chatbot for explorer yachts — because nothing says "smooth sailing" quite like talking to a robot!
P.S. This blog was started on Grok 3 (Xai), progressed on Claude (Anthropic), and was completed using Grammarly (Chat GPT). Salvation for a semi-literate sailor. Special thanks to John Johnson (Yachtbuoy) for access to extensive yacht video transcripts for AI training.
What Is a Chatbot?
A Chatbot is an AI application that engages in human-like conversations. Think of it as a virtual first mate who never gets seasick, doesn't drink all your beer, and instantly responds to inquiries about boats and maintenance—all without human intervention. Unlike static FAQs, chatbots provide dynamic, personalized responses based on your questions.
How It Was Created
We used natural language processing technology, allowing it to understand human interaction from both written and visual sources. Language with nuances and occasionally colorful sailor vocabulary.
The training process involved feeding the system targeted Explorer Yacht-related materials—forcing it to binge-read more boating content than a cabin full of "Yachting Monthly." We then refined this knowledge base as questions arrived by adding alternative meanings, categories, or flagging unanswered with additional training. Hopefully, it wouldn't tell someone to patch a hull breach with duct tape, but that remains a possibility.
How this AI Develops
It learns and improves over time—becoming the marine equivalent of that friend who started as a landlubber but now won't stop talking about jibing techniques.
As users interact with the system, it analyzes these conversations to identify patterns and information gaps. With each interaction, it should become more adept at providing comprehensive answers, gradually transforming into that knowledgeable dock neighbor who always has a useful solution (without unsolicited advice about your docking technique).
Well, at least that's the theory. So here are some example questions and answers from our testing, printed exactly as they landed. I'd give it about 80% accuracy, with some answers either failing or missing the point.
It did well with information it had been trained on (e.g., the Operations Manual for Vanguard). It also provided further reading and source links, which was a plus.
It did less well when requested to perform with more functionality typically associated with a larger AI Model - such as preparing a comparative performance table.
It also identified some incorrect older data (like a range of 7000 miles when it is in reality 2500/3000 at an acceptable speed). We can intervene here to retrain it based on real-world experience with performance and range data.
One feature of these Chatbots is that they can be continually trained and learn from the questions asked. So, ask away and have a play with the system! Once we have user data to analyse, I'll report and see how it improves.
We are returning to Vanguard tomorrow and will host a week-long visit from Praxis Automation to complete the commissioning of our control systems and iron out some evident bugs. More on that next Saturday before we start our run up the US East Coast to points further north.
Until then, have a good weekend.
Chris Leigh-Jones