Try the 8 test questions below. The bot keeps chat history only to resolve follow-ups like “what about in Turkey?”
The 8 test questions
Full written answers to all 8 are in test-results.md in the repo, and in the submission document.
Sources — only these 7 pages
How grounding works (read this before you test it)
1) Ingest: ingest.py re-fetches only the 7 URLs above (~1 request/second, clear User-Agent) and saves a clean text snapshot. 2) Knowledge base: knowledge-base.json holds short chunks, each checked against the page text, with its source URL. 3) Retrieve: the bot tokenises your question, scores chunks by keyword / phrase overlap (with a small synonym map, e.g. cost = price), and only answers from the best chunk(s) above a threshold. 4) Guardrails first: booking, weather, personal medical advice, and prices not in the pages are handled by fixed rules before retrieval — they can never be “answered” by inventing something. 5) Refresh: re-run ingest.py, diff the snapshots, update knowledge-base.json, re-run the 8 tests, update the “Last updated” stamp.
What this bot cannot do — honestly
It cannot book appointments, give a diagnosis or say whether a treatment is safe for you, quote a price that is not on these 7 pages, compare a price in a country that is not listed for that treatment, or describe a video beyond its title (no transcripts were in the extracted page content). For any of those, it points you to a free PlacidWay quote instead.
Get a free quote — lead capture
If you want a personalised quote, leave your name and email. This demo stores it only in your browser (localStorage) — nothing is sent to a server by this page. Use the button to open a pre-addressed email to PlacidWay if you want to send it.
Questions the bot could not answer
Logged locally in your browser only, so PlacidWay can find content gaps. 0 unanswered so far.
Medical note: PlacidWay provides medical travel information, not healthcare services. Prices on its site are estimates only — always confirm actual prices directly with the provider before booking, and consult your local licensed healthcare provider before pursuing any treatment.
Code & README: see the GitHub repo linked in the submission. Built by Antoinne Stewart for the PlacidWay AI Automation Engineer assessment. No secrets in the code; no API key is needed to run it.