Martin Michaux

AI engineer, builder, founder · Head of AI at Herald · Forbes 30 Under 30

An AI knowledge assistant that answers from your own documents

Your team keeps asking the same questions: how do we handle this case, where is that template, what did we agree with this client? I build assistants that answer from your own procedures, documents and history, show their sources, and say so when they do not know.

Background

From research to production

My MSc thesis at TU Delft, KarGus, was a knowledge-graph system for answering questions across many documents. Today I lead AI at Herald, whose AI agents work with a firm’s own placement data.

  • Answers grounded in your documents, not the model’s memory
  • Sources shown for every answer
  • Access rights respected, so people only see what they may see
  • Says “I don’t know” instead of guessing

What it can answer

Procedures and policies

How do we handle a refund, a complaint, a new client?

Products and pricing

Which option fits this customer, and what does it cost?

Past cases and projects

Have we done this before, and what did we decide?

Contracts and agreements

What did we agree with this client or supplier?

Technical documentation

Manuals, specs and internal wikis, searchable in plain language.

Onboarding

The hundred small questions every new team member has.

Where it lives

In Slack or Teams

Ask a question in the chat tool your team already uses.

On your intranet or a web app

A simple, private search-and-answer page for your team.

Inside your other AI workflows

Other agents use the same knowledge, for example to draft replies that follow your policies.

How I usually work

  1. 1

    Understand the work

    Look at how the task is done today, how often, by whom, and what a good result looks like. Often a simple automation is enough and no AI is needed.

  2. 2

    Pick one workflow

    Choose the task with the most repetitive hours and a clear definition of done, and measure it before changing anything.

  3. 3

    Build and test on real cases

    Build the workflow on real data and tools. At first a person approves the AI’s work, so nothing goes out unchecked.

  4. 4

    Measure, then expand

    Compare before and after. Let it run where it has proven reliable, then move on to the next workflow.

Frequently asked questions

What is a RAG assistant?

RAG (retrieval-augmented generation) means the assistant first searches your documents for the relevant passages, then writes an answer from them. It keeps answers grounded in your content and lets it show its sources.

Can it see confidential documents?

Only what you connect, and access can follow your existing permissions, so people only get answers from documents they are allowed to see.

What if the answer is not in our documents?

It says so, instead of making something up, and can point the person to the right colleague.

Which sources can it use?

Google Drive, SharePoint, Notion, Confluence, PDFs, websites, and data from your other tools.

Is my data safe? What about GDPR?

It should be designed with GDPR in mind from the start: data kept in the EU where possible, access limited to what the AI needs, and a person approving sensitive actions. The setup also needs documenting for your privacy records.

Related articles

How to implement AI in your business: a 6-step playbook for SMEs

Article · Playbook · 5 min read

AI agent, automation or chatbot: which one does your business need?

Article · Fundamentals · 4 min read

Thinking about AI for your company, or working on something similar? I’m happy to talk it through. You can send me an email or book a call.

Book a call with Martin