🧳 Tourist AI Bot

ℹ️ What is this tool?

In one sentence: a robot pretends to be the customer, asks our Ferry Chatbot questions, and then gives it a school report with marks out of 5.

🎯 Purpose

We need to know whether our Ferry Chatbot is actually good at its job. This tool plays the role of the tourist: it asks the Ferry Chatbot the customer's questions one after another, then marks every answer against the standards you set below. Just like a real customer, it cannot see our database — it only knows its own booking.

📋 How to use it (4 steps)

  1. Write down what a good answer looks like for each situation — cancellation, date change, refund, and so on (the box below).
  2. Pick the AI provider and model, or simply leave the default.
  3. Upload the transcript of a real call (required) and pick a difficulty for how the tourist speaks (clear, simple, broken English, garbled voice…). Only the tourist's questions are used.
  4. Press Run Simulation, then read the auto-detected goal, the marks, and the final report card.

⚙️ What happens in the background

  1. It reads the uploaded transcript and removes the travel operator's replies — only the tourist's questions are kept.
  2. It becomes the tourist — it knows only the details of its own booking and can never see our database, exactly like a real customer.
  3. It asks the Ferry Chatbot those questions one by one, in the order the customer asked them, waiting for each reply before sending the next.
  4. The Ferry Chatbot looks up real data to answer — it decides by itself which systems to check (booking lookup, ferry timetable, cancellation rules…) and builds its reply from what it finds. Any reply that used real data is marked in red, with a note of which system was checked and what it found.
  5. It judges every reply against the prompt you wrote — the expected answer for that situation (cancellation, amendment, refund…) — giving a score, a one-line reason, and flags such as "vague", "invented information" or "ignored my booking".
  6. At the end it writes a report card covering: (a) the overall score and a pass/fail verdict, (b) how many questions were answered well and which were not, (c) the strengths, (d) the weaknesses, and (e) suggestions on how to improve the Ferry Chatbot.

Demo only — nothing is saved.

Tourist AI Bot instructions (what to expect per scenario)

This is where you tell the Tourist AI Bot what a good answer looks like for each type of request. The bot uses these expectations to grade the Ferry Chatbot. Editable any time — changes apply on the next run.

Run

Choose the AI provider and model, and a difficulty for how the tourist speaks. Paste a Gemini transcript below — it is required; only the tourist's questions are used (the operator's replies are ignored). The bot works out the customer's goal from the transcript and keeps asking follow-ups until it's met.