Solution

RAG chatbot grounded in your own documents

A RAG chatbot answers from your documents. We index your manuals, PDFs and videos, retrieve the passages that match a question and answer from them. When the assistant is not sure, a person takes over with the whole conversation. One assistant serves web, WhatsApp and email.

What it does

  • Indexes the manuals, PDFs, videos and other files you already have
  • Answers customers on your website, on WhatsApp and by email from one knowledge base
  • Retrieves with reranking and a confidence threshold, and hands over below it
  • Passes the conversation to a person with the full context
For a German home-security manufacturer: 112 documents and videos indexed, and a 22-case evaluation suite that runs before every release.

How we build it

  • Audit the knowledge base. We check what is in it, what is missing and what contradicts itself, because the assistant can only be as good as its sources.
  • Rerank and threshold the retrieval. A weak match leads to a handover, not a guess.
  • Run an evaluation suite before every release, so a change that makes answers worse does not go live.
  • Replay real escalations. We replay real past escalations against the guardrails before each release.

What you get

  • An assistant on the channels you choose, answering from one knowledge base
  • A knowledge base audit with a list of gaps and conflicts
  • Handover rules, with the full conversation passed to your team
  • An evaluation suite you can re-run after every change

Related

Frequently asked questions

What is a RAG chatbot?

RAG stands for retrieval-augmented generation. The chatbot first searches your documents for the passages that match the question, then writes the answer from those passages. That keeps answers tied to your manuals and price lists instead of the model’s general knowledge.

What is the difference between RAG and a normal chatbot?

A scripted chatbot follows fixed paths. A plain language-model chatbot answers from general training and can sound right while being wrong. A RAG chatbot answers from your own files, so you can check an answer against its source. It still needs rules for when to stop and hand over.

How do you keep a RAG chatbot from making things up?

With three controls. Retrieval is reranked and has a confidence threshold, so a weak match leads to a handover instead of a guess. An evaluation suite runs before every release and blocks changes that make answers worse. And we replay real past escalations against the guardrails before each release.

Can it work on WhatsApp?

Yes. The same assistant answers on WhatsApp, on your website and by email, from one knowledge base. On WhatsApp we use the official WhatsApp Business API. A person can take over any conversation, with the full history.

Where is our data stored?

That is agreed per client. Options include hosting in the EU and a private deployment on your own infrastructure when documents cannot leave it. We document which provider processes what, and we sign data processing agreements where needed. This is not legal advice.

Book your free automation audit

Tell us what slows your team down. You get a clear picture of what to automate, what it saves and what it takes, no obligation.

Request the audit

We reply within one business day.

Prefer to see it live?

Message us on WhatsApp and experience an AI agent first-hand, the same technology we deploy for customer conversations. No forms, no slides, just a live demo in your pocket.

Get a WhatsApp demo

Or email us directly: [email protected]