RAG
Retrieval-Augmented Generation
An AI pattern where the model retrieves relevant source documents first, then answers using only those documents — instead of guessing from training data.
Large language models are good at fluent writing and bad at remembering specific facts. RAG fixes that by indexing a corpus of trusted documents — manuals, FAQs, specs, policies — and pulling the most relevant passages into the prompt at answer time. The model then answers using those passages and cites them.
For product support, RAG means an AI assistant that answers 'what's the torque setting on this bolt?' from the actual manual for that exact product — not a plausible-sounding hallucination based on similar products it saw in training.
RAG is most of what makes 'AI for product support' actually trustworthy. The hard part isn't the model; it's having a clean, structured, per-product corpus to retrieve from.