Sources
Which documents, systems and formats? DMS, drives, wiki, email, specialist systems.
RAG Engineering for reliable access to company knowledge.
RAG means a system searches approved documents for relevant passages and uses them to draft an answer with sources. Search, permissions and quality must be checked.
In brief: A RAG system searches documents for relevant passages and drafts an answer from them. The answer is meant to cite sources. Quality depends on sources, search, permissions and measurement.
Architecture
Six building blocks of a knowledge system.
Preparation
Split, structure, add metadata and permissions.
Search index
Semantic and keyword search across prepared text passages.
Search
Find relevant passages with a permission check for each user.
Draft answer
The model writes the answer from the passages found and is meant to cite them.
Measurement
Retrieval quality, answer quality, coverage – continuously in Studio.
Limits
Where RAG is meant to help and where its limits are.
| Intended support | Limit |
|---|---|
| Prepare answers from existing documents | Reliably fill in missing knowledge |
| Cite sources and respect permissions | Resolve contradictions in the sources on its own |
| Stay up to date when sources are maintained | Recognise outdated documents if nobody flags them |
| Run locally (Private AI) | Deliver good results without preparation |
Permissions
Access by role: who sees which answer.
A knowledge system must respect the permissions of its sources. That is why the search is designed to check each user’s rights before a text passage is passed to the model.
This applies to client data at accounting firms just as much as to project documents in construction.
Further reading
Related topics.
Frequently asked questions
Frequently asked questions about RAG Engineering
Is a chat with document upload enough?
For a test, yes. For live operation, it lacks permissions, up-to-date content, measurement and integration – which is exactly what RAG Engineering covers.
Does the model have to learn our documents?
No. RAG accesses the documents at runtime; the model is not trained. Fine-tuning is the exception, not the norm.
Do you show internal details?
Publicly, we show the principle, layers and data flow at a high level. Internal topology, routing rules, security rules and version states stay internal.
Is it already in production?
We state the maturity of each building block openly. Phrases such as “is designed to” describe the architecture, not a client deployment. Studio, Agents and parts of Factory run in our own development and operating environment and are continuously extended. Further operating models, multiple nodes and private and local installations will follow step by step.
Next step
The knowledge you want to make findable.
We check sources, permissions and operation for your case.