RAG (Retrieval-Augmented Generation)
Generation grounded by retrieving from a defined document set, rather than relying only on a model's training data.
RAG is a technique where a model's output is grounded by retrieving relevant passages from a defined, external document set at generation time, instead of relying solely on whatever it absorbed during training. The model still does the writing, but it writes with specific retrieved material in front of it, which is what gives the output something concrete to trace back to.
The reason this matters specifically for pharma content is that a model's training data has no idea what a particular product's current approved label or claim set says. SwishX's context-harnessed stage works on the same underlying principle: a request's context is merged with the brand dossier, the account's defined set of approved claims and sources, before generation happens. Every claim the resulting prompt implies is then checked against that same set, which is what makes a claim traceable to a source instead of to a general pattern the model learned elsewhere.