Whilst it's access to a wide range of knowledge is one of AI's core strengths, sometimes it's more appropriate to have your results be scoped to a more restricted set of reference material.
This pattern allows users to provides primary references for the AI to reference in it's response.
All users will have an intuitive view of what material they'd like to the AI to focus on. However, as a general understanding of how AI's (LLM's) actually work grows, it will take some time before most users comprehend the difference between training data, publically accessible data & reference material.
Consider your users understanding of these concepts to ensure they can utilise scoped reference material effectively.
The key variables for this pattern are simply how many, how large & what file types is the reference data you allow your users to add. These choices are highly contextual to your users use case. You may want to limit it to only a few references or allow for a complex multi format reference set to be authored.