Hi @Will_Maggi
It possible to do that by building a RAG(Retrieval Augmented Generation) pipeline where the semantically similar information can be retrieved and supplied to model as context to reply to users’ queries.
To run the retrieval search, embeddings can be used.
The context length of the Gemini Pro 1.5 is around a million Tokens which means you can even supply quite large pieces of text as context, which is simpler to build and test but can add to cost per run.