RAG:AI that knowsyour knowledge

We connect language models to your documents, files and databases. The AI then answers from your knowledge and shows where each answer comes from.

Discuss a RAG project

On this page

Enquiry

Nikolai Schöbel und Jeremias Burger, Co-Founder Scalableloops

Let's talk about your project.

First we check whether the project fits your business model. Then you get a proposal with phases and effort.

Discuss a RAG project or call: +49 151 1576 5566

RAG pipelines

Retrieval augmented generation (RAG) means that before a language model answers, a retrieval step finds the relevant passages in your own material and passes them to the model. The AI then answers not from its general training knowledge but from your contracts, manuals, case files or product data, and can name the source. Scalableloops builds such RAG pipelines and knowledge graphs for companies and organisations whose knowledge is scattered, extensive and often confidential, for example in law, public administration and healthcare.

In brief

RAG connects a language model to a search over your own documents.
The term comes from a paper by Patrick Lewis and colleagues, presented at NeurIPS 2020.
A knowledge graph adds relationships: who, what, responsible for which task, valid since when.
For confidential data we plan hosting in Germany or operation on your own servers.

As of 26 Sep 2026Scalableloops GmbH, Eggenfelden7 min read

Nikolai Schöbel and Jeremias Burger

Co-founders of Scalableloops GmbH. Nikolai Schöbel leads online marketing and AI strategy, Jeremias Burger the AI architecture. Both build AI systems and train teams on them in their own agency work.

On this page

What is retrieval augmented generation?
When do you also need a knowledge graph?
Where do we use RAG and knowledge graphs?
What can RAG not do?
Frequently asked questions
How we connect AI to your knowledge
Where the information on this page comes from

Definition

What is retrieval augmented generation?

Large language models store knowledge in their parameters, in what they learned during training. That knowledge is general, has a cut-off date and does not include your internal documents. This is where RAG comes in.

The term goes back to the paper “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” by Patrick Lewis and colleagues, accepted at the NeurIPS conference in 2020. The authors combine the knowledge stored in the model with a searchable collection of text that the model draws on when answering. The principle is the same today: retrieve first, then generate.

For your business this means the AI receives, for each question, the few paragraphs from your material that fit, and bases its answer on them. Because the passages are known, every answer can be checked against its source. For implementation, most of the work is not in the model but in making the right information findable.

Knowledge graph

When do you also need a knowledge graph?

Plain text search finds similar wording. It does not know that contract A was amended by addendum B, that Ms X is responsible for site Y, or that a rule only applies from a certain date. A knowledge graph captures such relationships: it stores things and their connections, not just text.

Google popularised the idea in 2012 with its Knowledge Graph under the motto “things, not strings”. In companies we use it where questions must be answered across several documents, for example in case files, rulebooks and responsibilities.

Question	Text retrieval (RAG)	With knowledge graph
“What does the manual say about returns?”	fits	not needed
“Which version of this contract applies today?”	uncertain	fits
“Who is responsible for this case?”	uncertain	fits
“Summarise all minutes on project X.”	fits	helps with assignment

In practice

Where do we use RAG and knowledge graphs?

At JourF’x we are responsible for the AI pipelines that understand domain context. Conversations in medicine, law and public administration become structured, audit-proof documentation there, supported by the material that belongs to the appointment.

For Defence:Connect we deliver AI pipelines that make rulebooks such as EN 9100 and export control requirements usable for mid-sized manufacturers, with traceability as a precondition. For security-critical areas we also plan with models that run on your own infrastructure.

How we approach local operation and what “local” really means technically is covered in Local AI for business.

Limits

What can RAG not do?

RAG is only as good as the material it draws on. Outdated, contradictory or incomplete documents lead to answers of the same kind. Every project therefore starts with a review of the sources: what is current, what is duplicated, who maintains it?

And: an answer with a source can be checked, but it is not automatically correct. Decisions with legal or medical consequences stay with a person.

Frequently asked questions

What is the difference between RAG and fine-tuning a model?

Fine-tuning changes the model itself, which is costly and hard to reverse. With RAG the model stays unchanged; for each question it receives the relevant passages from your material. New documents are usable right away.

Can RAG work with confidential data?

Yes, if storage and model fit the requirements. Depending on how sensitive the data is, we plan hosting in Germany or operation on your own servers.

Which documents can be included?

Typically manuals, contracts, minutes, knowledge articles, product data and database records. Which sources make sense is clarified in the initial review.

Do I always need a knowledge graph?

No. For questions that one document can answer, good retrieval is enough. A knowledge graph pays off when relationships, versions or responsibilities across many documents matter.

How much does a RAG project cost?

It depends on the volume and type of sources, the connected systems and the requirements for operation. After a conversation you receive a proposal with phases and effort.

Process

How we connect AI to your knowledge

01
Collect questions
Which questions does your team ask again and again, and where are the answers?
02
Review sources
We check your material for currency, duplicates and ownership.
03
Pilot
The AI answers a defined set of questions with sources while your team rates the answers.
04
Operation
We define how new documents are added and how quality is checked on an ongoing basis.