Frequently asked questions about internal documentation, RAG and business knowledge
26 questions
Internal documentation → Can AI search internal documentation?
Yes. It can use manuals, procedures, regulations, knowledge bases and other sources to answer in natural language.
What is RAG?
RAG is an architecture that retrieves relevant information before asking the model to generate an answer.
How is RAG different from training a model?
RAG uses external information at the moment it answers. Training or fine-tuning a model changes its internal behavior.
What is a vector database?
It’s a technology that stores semantic representations and retrieves related content by meaning.
Does RAG always need a vector database?
No. It can combine semantic search, filters, metadata, traditional search, databases and other mechanisms.
Can the system cite its sources?
Yes. It can return the documents, excerpts or references it used to generate the answer.
Can it respect per-user permissions?
Yes. Retrieval can be limited to the sources each user, role or department is authorized to see.
What happens when a document changes?
The new or changed document is processed again and replaces the old version. Before any answer changes in production, it goes through the test suite.
Can it work with Google Drive?
Yes, as long as there’s a compatible integration and sufficient permissions.
Can it work with PDFs?
Yes. It can also use other formats depending on the project’s needs.
Can it work with sensitive documentation?
Yes, but permissions, infrastructure and suitable providers have to be defined.
How do you reduce the risk of hallucinations?
By limiting answers to retrieved sources, evaluating retrieval and defining what should happen when information is missing.
Can the system say it doesn’t know an answer?
Yes. In fact, in some contexts it’s better to admit missing information than to generate incorrect content.
What types of documents can a RAG system search?
PDFs, text documents, internal pages, wikis, knowledge bases and other formats that can be processed properly.
Can it combine documents and databases?
Yes. A system can retrieve structured and unstructured information in the same query.
Can it use metadata to improve answers?
Yes. Date, department, document type, version or category can be used to filter and prioritize results.
Can it tell current documents from outdated ones?
Yes, if there’s metadata or rules that identify the active versions.
What is hybrid search?
It’s the combination of semantic search with keyword search or other retrieval methods.
Can a RAG system give a wrong answer even if it finds the right document?
Yes. That’s why retrieval and generation have to be evaluated separately.
How do you evaluate a RAG system?
By checking whether it retrieves the right sources, whether it answers faithfully and whether it cites properly.
Can answers be limited to internal documents only?
Yes. It can be set up so it doesn’t use outside knowledge when the query calls for company sources only.
Can it show several sources for one answer?
Yes. It can cite one or more documents when the answer depends on different sources.
Can it use a hierarchy of sources?
Yes. Some sources can take priority over others based on their authority or how current they are.
Can it detect that two documents contradict each other?
It can spot discrepancies, although deciding which one prevails requires suitable rules or metadata.
Can a documentation system learn from new questions?
It can use real queries to find content gaps and improve retrieval, although it shouldn’t retrain itself without control.
Tell us about the process you want to improve.
You do not need to know which model to use, which architecture you need or whether the problem calls for an AI agent. Tell us how the process works today, who is involved and which tools you use.
We’ll reply within one business day with a first read of your case.