RAG Systems
Your company knowledge. Securely accessible and traceable.
Your team needs answers from the right documents. A RAG system first searches approved company information, then uses it to compose answers with source references. You decide who may use which knowledge.
The introductory call covers your use case, available sources and protection requirements. This helps establish whether RAG or a simpler solution is appropriate.
BUSINESS PROBLEM
Knowledge is scattered. Answers take too long.
A colleague needs the current procedure, a service team needs a product detail, or a new employee needs an internal answer. When the relevant information is spread across PDFs, folders, CRM and ERP, the task becomes finding the right source. We start with recurring questions like these and define which information each role may use.
Only approved sources may form the basis.
Answers to recurring questions should be easy to find without copying information between tools.
If the source is missing, the system escalates instead of guessing.
FIT
When RAG helps — and when it does not
RAG is often useful
- Many authoritative documents that come up regularly
- Answers must be grounded in approved sources and constrained by access rules
- Roles decide who may see which knowledge
A simpler solution may be better
- A few stable FAQs or a clear process handbook
- Rare one-off cases without a recurring pattern
- Decisions that always need a person for legal or financial reasons
FLOW
From source to a controlled answer
The work starts with your team’s language: which questions recur, what counts as a reliable answer, and where the approved information lives. We shape the source structure and answer format around that use case, then implement retrieval and access rules.
Only content you have explicitly released.
The system finds the relevant passages for the question.
The answer stays bound to the retrieved context.
It is clear where the statement comes from — or the system refuses.
CONTROL
Who can see what — and who reviews uncertain answers?
Permissions determine which sources a person may use before an answer is composed.
If a suitable source is missing, the system says it does not know the answer.
Sensitive or unclear cases go to a responsible person.
Only necessary information is processed or stored.
QUALITY ASSURANCE
A plausible answer is not enough.
A RAG system is useful only when it retrieves relevant knowledge, keeps claims grounded in approved sources, respects access rules, and refuses or escalates when confidence is insufficient.
Retrieval quality
Does the system find the relevant, approved passages for the question?
Source grounding
Can the answer’s claims be supported by the retrieved context?
Access and boundaries
Do roles, permissions, refusal behaviour and human escalation work as intended?
Operations and security
Are data flows, sensitive information, prompt-injection risks, logging and deletion handled transparently?
Automated metrics provide signals, not a final quality verdict. Critical cases are checked against subject-matter-approved reference questions and reviewed by people.
Three examples for acceptance testing
Designed scenarios explaining the method. They show the intended behaviour; no test using client data has been performed here.
The source is approved
Question: “Who approves an expense claim?” The answer identifies the responsible role and references the relevant passage in the current expense policy. A subject-matter reviewer checks both the claim and the source.
The information is missing
Question: “Does this also apply to the new location?” If the approved documents do not say, the system should identify the gap and refer the question to the responsible person.
Access is restricted
A request for another employee’s payslip must not reveal the document, confidential passages or amounts. This scenario is checked using different roles.
DATA FLOW
Know where your information is processed.
We assess hosting in Switzerland or on your own infrastructure against the sources, integrations and protection requirements.
Which sources, which processing, which provider: named before work starts.
We check whether a provider may use data for model training against its terms, the agreement and the selected settings.
We work with privacy in mind. This is not legal advice.
WORKING TOGETHER
From a suitable use case to a checked handover.
The introductory call establishes suitability. The individual proposal then defines what will be built and the criteria for accepting it.
For teams with recurring questions about approved company knowledge. We agree sources, roles, integrations and a bounded first use case.
You receive the agreed knowledge search with an answer interface, source references and access rules. Additional integrations and operational services are specified in the proposal.
Your subject expert checks agreed reference questions, missing sources and denied access. We record results, unresolved defects and the release decision.
Documentation covers source maintenance, permissions and known limits. Ownership, support and later extensions are explicitly agreed.
FAQ
Definition and limits
What is a RAG system?
RAG first retrieves approved company information and then forms an answer. That makes answers more traceable and easier to constrain. AI output remains probabilistic and needs human review for important decisions.
When is RAG useful?
When employees repeatedly search documents, folders, CRM or ERP for authoritative answers, and those answers must be grounded in approved sources and constrained by access rules.
When is a simpler solution better?
When the knowledge base is small, questions are infrequent, or the answers already exist in a clear FAQ, intranet or process handbook. In those cases, better structure is often enough — without a retrieval system.
What happens if the system does not know the answer?
It should refuse or say “I do not know” and escalate to a responsible person. The system must not guess in unclear or sensitive cases.
Does all data stay in Switzerland automatically?
No. Swiss hosting or on-premise are architecture choices based on the required protection level. Data flows are documented. This is not a residency or compliance guarantee.
Does this automatically mean no-training?
No. No-training language applies only where the selected provider and agreement actually support it. We review that before use and do not invent a blanket assurance.
What do we receive from the introductory call?
We discuss a typical question, suitable sources and the permissions required. You receive an initial assessment and know which information is still needed for an individual proposal.
Quoted after suitability
RAG systems do not have a publicly stated fixed or entry price. Suitability, sources, roles, integrations and protection requirements are clarified first. The exact scope is then quoted individually. Public visibility in search engines and AI answers belongs to our SEO/GEO work.
Request a RAG introductory callNEXT STEP
A RAG conversation or SOURCE/01
We clarify whether a controlled knowledge system is the right next step.