Lesson 8 of 10
Prompt design, structured output, and RAG
A useful prompt states the task, gives relevant context, sets constraints, and describes the expected output. Ask for a format that your application can validate, and handle invalid or missing fields instead of trusting a model response blindly.
Task: Summarize the supplied support note.
Rules: Use only details in the note. Do not invent dates or promises.
Output: JSON with keys "summary" and "follow_up".
If the note does not answer a question, say "not provided".Retrieval-augmented generation (RAG) searches an approved document collection and supplies relevant passages as context for a model response. A RAG system still needs source permissions, careful retrieval, citations, and checks that the answer is actually supported by the retrieved text. It does not guarantee correctness.
Try it: Improve a vague prompt such as “tell me about this report” by specifying the audience, source limits, length, and output format.