A team wants a Foundry-hosted chatbot to answer questions grounded in the company's internal product manuals rather than the model's general training data. To set this up as a retrieval-augmented generation solution in Azure AI Foundry, which two components must they configure? (Select TWO.)
Select all correct answers, then click Submit.
Short Explanation
Getting a chatbot to actually answer from the company's own manuals instead of guessing from general training takes two pieces working together: something that connects the assistant to where those manuals actually live, and something that turns their content into a form the assistant can quickly search through and retrieve relevant passages from, like a searchable index built on top of the source material. Skip either one and there's nothing to ground the answer in. Retraining the model's weights on the manuals is a different approach entirely, one that bakes knowledge into the model itself rather than letting it look things up fresh each time, so it doesn't provide the connection or searchable structure needed here. And a filter that blocks technical language would actively work against the goal, since the whole point is to let the assistant discuss the manuals, not censor the vocabulary they're written in.
Full Explanation
The correct answers are C and D. A retrieval-augmented generation setup needs a connection to a data source so Foundry can access the product manuals, and a vector index built from embeddings of that content so the assistant can retrieve the most relevant passages at query time and ground its answers in them; both pieces are required for grounding to work at all. Option A is incorrect because fine-tuning retrains the model's weights on a training set, which is a separate customization technique from retrieval and does not, by itself, connect the assistant to the current manuals or let it cite them at query time. Option B is incorrect because a content safety filter screens generated or input content for policy violations such as hate speech or violence; blocking technical vocabulary would actually prevent the assistant from discussing the manuals at all and has nothing to do with grounding answers in company data.