Azure RAG Patterns: Azure AI Search & OpenAI Pipelines

This course contains the use of artificial intelligence.

Build a production-minded Retrieval Augmented Generation application on Azure instead of stopping at a toy chatbot.

This hands-on course takes you from Azure resource foundation to secure document ingestion, hybrid retrieval, cited chat, evaluation, and release review. You will use Microsoft Foundry, Azure AI Search, Azure OpenAI models, managed identity concepts, and practical engineering templates that you can adapt to an internal knowledge assistant.

Who this course is for

•Azure developers building enterprise AI applications.

•AI engineers and data engineers responsible for search, indexing, or model integration.

•Solution architects designing secure RAG and chat systems.

•Technical leads who need an evaluation and production-readiness approach.

•Developers who understand basic Python or REST APIs and want an end-to-end Azure implementation.

What you will learn

•Design a complete retrieve–augment–generate architecture on Azure.

•Map Microsoft Foundry, Azure AI Search, model deployments, storage, and identity to clear responsibilities.

•Prepare, chunk, enrich, embed, and index enterprise documents.

•Choose between keyword, vector, hybrid, semantic, and agentic retrieval patterns.

•Build a chat orchestration flow with bounded context and source citations.

•Apply retrieval-time security filters and minimize sensitive data exposure.

•Write a grounded-answer and refusal contract for the model.

•Test for prompt injection in retrieved documents and user queries.

•Evaluate retrieval relevance, groundedness, citation coverage, latency, and cost.

•Use a release checklist to decide what must change before production.

Requirements

Students should have an active Azure subscription or access to a shared lab environment, basic familiarity with Azure resources, and basic Python or REST knowledge. No machine learning training or fine-tuning experience is required. Students should verify service availability, quotas, model deployment access, and pricing before creating billable resources.

Final project

You will build a cited internal policy assistant. The application ingests a controlled document set, stores structured chunks in Azure AI Search, retrieves evidence with hybrid search, sends bounded context to a deployed model through a Microsoft Foundry-connected workflow, and returns an answer with source references. You will then run a golden-question set, test an out-of-scope question and a prompt-injection document, record quality and latency results, and complete the production-readiness checklist.