AI that answers from your company data
Retrieval-Augmented Generation (RAG) — a system that "reads" your documents, contracts and manuals and answers like an experienced colleague. With source citations.
How does RAG work?
RAG combines a language model with search across your documents. It can show the sources behind an answer and reduce errors, but cannot eliminate them. I test quality against questions from your team.
PDF, Word, Excel, web pages, e-mails, Confluence, SharePoint — RAG processes any text format.
The system splits documents into semantic chunks and builds a vector index for search by meaning, not just keywords.
Users ask in natural language. RAG finds the relevant passages and produces an accurate answer with source citations.
Where you'll use RAG
Employees ask the AI instead of digging through dozens of folders and documents. Answers in seconds instead of minutes of searching.
"What are the termination terms in the contract with supplier X?" — RAG finds the exact clause and points to the paragraph.
Find relevant passages on GDPR, ISO standards and internal policies. How current the answers are depends on source updates and re-indexing.
Developers, engineers and support teams ask the system instead of searching the wiki. RAG understands technical language and context.
New employees learn faster — they ask the AI anything from company materials and get clear answers.
RAG combined with an AI agent — the agent answers customers exactly according to your documentation, pricing and terms.
What you get
The technical solution
- A RAG system with an index of your documents
- A web interface or API for queries
- Source citations with every answer
- Access control (who sees which documents)
- Automatic re-indexing when documents change
The results
- Faster search with links to source documents
- Accuracy measured on your real questions
- Time saved verified in a pilot
- A shared set of sources for the whole team
Sound interesting?
Describe your problem — I'll design a solution and calculate the ROI. The 30-minute call is free and non-binding.
Illustrative calculation: measuring RAG value
Imagine a team that currently spends eight hours a week searching documents. The figures below are assumptions; a pilot measures actual time and answer quality.
Frequently asked questions
Did not find your answer? Try the full FAQ.
It depends on the architecture. Data can be processed on your server or in a private cloud; an external AI service receives data according to the agreed configuration. Before launch, I map data flows, access rights and the required agreements, including a DPA.
PDF, Word, Excel, PowerPoint, HTML, Markdown, e-mails, CSV, TXT and more. We also connect to Confluence, SharePoint, Google Drive, Notion, OneDrive — wherever your data lives, it gets read.
For supported sources, I set up scheduled synchronisation or change notifications. A new version appears only after re-indexing; for important answers, users should check the date and cited document.
A simple RAG with no integrations (just PDF uploads) — 1–2 weeks. Advanced RAG connected to SharePoint/Confluence — 3–6 weeks. It depends on the number and quality of sources. We always start with a pilot on a smaller document set.
A hallucination is a plausible answer that lacks support in the sources. RAG supplies relevant passages and makes citations possible, but errors can still occur. I configure the system to say "I don't know" when evidence is insufficient and test it on your questions. Microsoft Foundry's documentation explains the limitation.
A basic knowledge base of a couple hundred documents sits in a different bracket from a company system with unlimited documents, role-based access control and automatic updates. I'll give you the price once we know how much data and how complex an access structure you need.
An internal knowledge system for employees usually doesn't fall under the stricter rules — it's not a high-risk system. If the RAG powers a customer-facing chatbot, the standard obligation to disclose it's AI applies (from 2 August 2026). Details in the AI Act article.
RAG is more than a chatbot on top of documents
If you're considering a knowledge system, it helps to first understand the difference between a regular AI agent, a chatbot, and a RAG solution connected to company data.
Read the related articleYour documents hold the answers — let AI find them
Tell me what documents and data you have. I'll design a RAG solution for your exact situation. The call is free and non-binding.