RAG
RAG is a method in which a language model, before answering, retrieves relevant passages from a business’s own documents and takes them as context. Beyond its training data, it answers from current, business-specific sources and can cite them. It lowers the risk of errors without removing it. Quality depends on tidy documents and accurate retrieval.
- Also called
- retrieval-augmented generation
- In Turkish
- RAG
- Read more · Service
- AI customer service assistant
Why it matters
A language model does not know the business’s products, pricing policy or internal rules. RAG grounds the answer in the business’s own documents. If the documents are outdated or scattered, the answers are poor too, so the source documents first need to be kept current and tidy.
Example
Say a firm sets up an assistant that answers customer questions based on its returns policy document. When a question arrives, the system finds the two relevant paragraphs and gives them to the model. The answer states, for example, “the return period is 14 days”, and shows its source.
Related terms
- HallucinationA hallucination is a language model stating information that does not exist or is wrong as if it were true, in a confident voice.
- Human-in-the-loopHuman-in-the-loop means leaving the decision or final check at a particular step of an automated process to a person.
- AI agentAn AI agent is a system built on a language model that picks its own steps towards a goal and uses tools to take them.
- APIAn API is the interface a piece of software offers to other software, defining which requests it answers and in what form.
- Error workflowAn error workflow is a separate workflow that runs when another one fails.
- IntegrationAn integration is a connection that lets two or more pieces of software pass data to each other without manual transfer: a site and a CRM, a shop and accounting, a form and an email tool.
The definitions are general information. If you think one is incomplete or wrong, write to us and we will reply within one working day. Get in touch
From our journal
From what we have written on Automation and AI; they show how the term turns up in the work.
- How to write prompts for work: 5 parts, an example, a checkThe five parts from role and context to example, a before and after on meeting notes, how to check the answer, what never goes into a public tool, and a shared prompt library for the team.
- What is RAG? AI that answers from your documentsMaking a language model search your own documents before it writes an answer: the parts of the system, where it goes wrong, testing with a fixed question set and personal data points.
- e-Fatura in North Cyprus: who must switch, how to applyWhat the Income and Tax Office’s pages and the Regulation it publishes say: paper invoices and e-Fatura stand side by side, who the obligation covers, the steps of the application, and what a business selling online has to connect.
- How to clean customer data before a CRM importTaking a copy, one row per person and a fixed set of columns, one format, duplicates, empty values, source and permission, a twenty-row test and the rules that keep the list clean.