WEBVTT

00:00:00.100 --> 00:00:06.387
This independent n8n demo shows a controlled knowledge-base assistant using synthetic data.

00:00:06.427 --> 00:00:08.890
A message arrives through a webhook.

00:00:09.450 --> 00:00:16.110
The workflow retrieves relevant clinic information and safety rules, then decides whether the request can be answered.

00:00:16.150 --> 00:00:21.560
A supported cleaning question returns a concise answer with a validated knowledge-base citation.

00:00:22.120 --> 00:00:28.697
The documents do not contain an implant price, so the workflow does not invent one and routes the request to a person.

00:00:28.737 --> 00:00:32.790
Emergency language is handled before generation, so the model is skipped,

00:00:33.250 --> 00:00:36.501
and the response directs the user to emergency services.

00:00:36.541 --> 00:00:40.497
All three paths were live-tested in n8n with a local language model.

00:00:40.537 --> 00:00:46.550
Bring your documents, rules, and preferred channel, and I will confirm the scope before implementation.