AI / Voice
Retrieval-Augmented Generation (RAG)
Retrieval-Augmented Generation (RAG) retrieves relevant information from approved sources and supplies it to a language model when generating an answer.
Retrieval-Augmented Generation (RAG) retrieves relevant information from approved sources and supplies it to a language model when generating an answer.
Retrieval-Augmented Generation, or RAG, is a method for giving a language model relevant source information at the time it answers a question. Instead of relying only on patterns learned during model training, the system searches an approved knowledge collection, retrieves matching passages, and adds them to the prompt used to generate the response.
How RAG works in a voice agent
The caller asks a question, speech-to-text produces a transcriptTranscriptA transcript is the written record generated from a spoken conversation, typically showing what the caller and agent said during a call., and the system searches connected content for relevant information. The retrieved text is supplied to the language model together with instructions and conversation context. The answer is then converted to speech. Thoughtly's Genius knowledge baseKnowledge baseA knowledge base is a maintained collection of approved business information that an agent can use to answer questions. supports text, PDF, audio, CSV, and URL sources, then connects selected databases to agents.
Practical use cases
- Answer approved questions about products, services, policies, or procedures.
- Give different agents access to a shared, maintained source of business information.
- Keep detailed reference material outside the main conversation prompt.
- Update source content without retraining the underlying language model.
RAG improves grounding, but source quality still decides the ceiling. Outdated pages, contradictory policies, poor formatting, and irrelevant retrieval can produce a wrong or confusing answer. Thoughtly's documentation recommends concise, current, natural-language content and notes that URL content is fetched as static content during setup.
Frequently asked questions
Does RAG train the language model on company data?
Not in the usual retrieval workflowWorkflowA workflow is a defined sequence of steps, decisions, actions, delays, and outcomes used to complete a business process.. It supplies selected information as context when the model answers, rather than retraining the base model.
Does RAG eliminate hallucinations?
No. It reduces reliance on unsupported memory, but teams still need clean sources, narrow instructions, testing, and escalationEscalationEscalation moves a conversation to a person, specialist, supervisor, or alternate workflow when the agent should not continue alone. for uncertainty.