Skip to content
George Wanjii
RAG chatbots and AI agents

Most bots guess.Yours will not.

RAG chatbots that answer from your own price lists, policies and FAQs. When the AI does not know, it says so.

You give it your documents. It reads them and answers the questions your team keeps answering by hand, day and night, without inventing anything you did not write down.

Why these go wrong

The problem is what you feed it

Almost every bad answer comes from the documents, not the software. Give it messy files and it will answer confidently from the wrong one.

It makes up a policy

A customer asks about refunds. The bot sounds completely sure of itself and promises them money back you never offered.

It reads the old price list

You changed your prices. The old list is still in there. The bot picks one of them, and it might not be the right one.

It will not say I do not know

Ask it something you never gave it an answer for and it answers anyway. Confidently. Wrongly.

In 2024 a car dealer's chatbot was talked into agreeing to sell a car for one dollar. Nobody had told it what it was not allowed to say.

How I build them

It admits when it does not know

01

It tells you when it does not know

If the answer is not in your documents, it says so and offers to pass the customer to a person. That is the whole trick, and most bots skip it.

02

It shows where the answer came from

Every reply points at the document it read. Your team can check it in one click instead of hoping it got it right.

03

Anything about money follows your rules

Prices, refunds, delivery times and anything touching an account come from fixed rules you set, not from the bot's opinion.

04

One set of files, no contradictions

Old versions and documents that disagree get cleaned up first. Most wrong answers start here, not in the software.

05

You see every question it could not answer

A list of what your customers asked and the bot could not handle. That list tells you exactly what to write next.

What you get

You get a working bot

  • A chatbot on your website, on WhatsApp, or both
  • Your documents loaded in, tidied up, old versions removed
  • Rules for what it must never answer on its own, agreed with you first
  • A simple screen showing what customers asked and what it said
  • A short guide so your team can add new documents without me

About three to four weeks from you sending your files to a bot you are happy to put in front of customers.

Before you ask

What I will not promise

Nobody can sell you a bot that is never wrong. Anyone who says they can is showing you a demo, not something that has met real customers.

What I can build is one that is wrong safely. It stops instead of guessing, it shows you what it read, and the answers that could cost you money follow your rules.

If your documents disagree with each other, week one is fixing that. It is boring work and it is the difference between a bot that helps and one you quietly turn off.

Tell me what it needs to answer

You do not need to know what to ask for. Tell me the questions your team answers most often and I will tell you if a bot helps.

Already have a bot behaving badly? I will look at it free.

RAG stands for retrieval augmented generation. It means the AI looks up your documents before it answers, instead of guessing from what it was trained on.