RAG stands for retrieval-augmented generation. An application finds relevant information in approved sources and supplies it to a language model alongside the question. The model then uses that context to compose an answer. It is useful when an answer depends on company information that the model would not otherwise have.
- 01Ask a question
- 02Check access
- 03Retrieve evidence
- 04Compose an answer
- 05Show sources
Illustrative process. The exact components and controls depend on the workload.
Follow one question through the system
Imagine an employee asking, “Which expenses need manager approval?” A useful assistant must know the employee’s business unit, retrieve the current expense policy and identify the applicable threshold. It should point to the policy section, rather than simply sound confident.
The work starts before the question: documents are extracted, divided into useful passages and indexed with source, version and access metadata. At query time, the application retrieves candidates, selects relevant passages and asks the model to answer from that evidence. Semantic search can help match meaning; keyword search remains useful for exact product codes and identifiers.
What companies can use it for
- Help service staff find troubleshooting steps across manuals and resolved tickets.
- Let employees search current HR and operational policies, with answers restricted to their access rights.
- Help sales teams locate approved product information and supporting proposal material.
These are examples of possible workflows, not claims about completed Novacom client projects. Choose a task with a clear source of truth and a person who owns the answer.
Where projects usually become difficult
Uploading every shared drive does not create a trustworthy assistant. Old policies may contradict new ones; scanned tables may extract badly; a source may contain instructions that should be treated as untrusted text. A citation proves that a passage was retrieved, not that it supports every sentence in the answer.
For the expense example, test two versions of the policy, an employee from another business unit and a question the documents cannot answer. The system should select the right version, respect permissions and state when evidence is missing. These tests tell you more than a polished demonstration.
A useful first pilot
Choose one document collection and collect real questions from its users. Record the expected answer and source for each question. Include missing answers, ambiguous wording and permission boundaries. Compare the assistant with the existing search process using task completion, incorrect answers and time spent verifying results.
Keep a feedback route to the document owner. Sometimes the right fix is a clearer policy, not a different model. Expand the collection only after you can explain why the first pilot succeeds and where it still fails.
Does RAG replace a database?
No. Use an authorised database or API for live balances, order status and other structured facts. RAG is especially useful for discovering and explaining document-based knowledge. Many applications combine both, while keeping their sources and permissions distinct.
Sources & further reading
Primary references for the technical background and regional statements in this guide. Planning examples and checklists are Novacom’s practical guidance; examples are illustrative unless explicitly identified as project experience.