Skip to content
Turing College Logo
Submit projectSign In

Ask the boat

Ask the boat

About this project

Asks the manual. Quotes it word for word. Shows the page. Never guesses a step.

The problem

A charter operator runs 10 to 40 boats, and every week new guests take them out. "How do I start the generator?" sits in a 200-page English manual nobody reads at anchor, so guests phone the base. A general chatbot is the wrong fix: it guesses, it cannot tell one boat's watermaker from another's, and it must never show a guest the owner's papers or the engineer's notes.

How it works

Ask the boat answers only from the manuals of the boat the guest is on. Every step in an answer is a quote taken from the manual by code, not written by the model, and each quote links to its page. Safety warnings come first. When the manuals do not cover a question, it says so and points to the base instead of inventing a step. Every passage carries a clearance, and row-level security in Postgres decides what a guest, a skipper or an engineer may retrieve. Guests also get a small assistant for their booking and for emergencies.

Built with

  • Python (FastAPI)

  • React and TypeScript

  • Postgres with row-level security

  • hosted writing models with zero data retention

  • a decision model for reranking and checks

Measured on 36 test questions: every quote identical to the manual, no access leaks; answer speed and strict faithfulness are still below the targets we set, and the README says so with the numbers.

Turing College, AI Engineering, capstone project.

Created byChristopher Bauer
Published atNovember 17, 2026
CourseAI engineering
Looking for
Freelance projectsPotential users
OpenOpen LinkedIn