Summary

Completed

Retrieval Augmented Generation connects your database to the capabilities of large language models. Instead of relying on a model's training data, you provide current, relevant information from your own tables.

The entire RAG pattern executes in T-SQL. Your database orchestrates the flow: search, format, prompt, call, parse. You can add AI capabilities to existing applications by modifying stored procedures, without rearchitecting your application stack.

In this module, you learned how to:

• Identify RAG use cases: Recognize scenarios where grounding Large Language Model (LLM) responses in database content improves accuracy and relevance
• Prepare context from SQL: Use FOR JSON to convert query results into text that LLMs can process effectively
• Construct augmented prompts: Build request payloads that combine system instructions, retrieved context, and user questions
• Execute the RAG pipeline: Call Azure OpenAI endpoints using sp_invoke_external_rest_endpoint and parse the responses

Learn more

• sp_invoke_external_rest_endpoint (Transact-SQL)
• Intelligent applications and AI
• Format query results as JSON with FOR JSON
• Intelligent applications and AI FAQ