LLM and GenAI App Deployment on Azure

Course Description

This course provides a comprehensive, production-focused journey into building and deploying Large Language Model (LLM) and Generative AI applications on Microsoft Azure. Starting from the architectural principles of agentic AI systems, learners progressively build expertise across the full deployment stack — from advanced prompting and multi-agent orchestration, to RAG pipelines, containerized infrastructure, and automated CI/CD pipelines.

Each module combines conceptual grounding with hands-on implementation, ensuring learners can not only understand how modern GenAI systems work but can configure, deploy, monitor, and maintain them in real production environments. The course concludes with a capstone project where learners ship a fully containerized, cloud-deployed GenAI application of their own.

Who Is This Course For

This course is designed for developers and engineers who have a foundational understanding of Python and cloud concepts and are ready to take the next step into production AI engineering. It is ideal for:

• Software and backend engineers looking to integrate LLMs into scalable applications

• Cloud and DevOps engineers who want to extend their expertise into AI infrastructure and MLOps

• ML/AI engineers seeking to operationalize models beyond notebooks and into real deployments

• Solutions architects designing enterprise-grade GenAI systems on Azure

• Full-stack developers building AI-powered products end-to-end

Course Objectives

By the end of this course, learners will be able to:

1. Explain the architecture of agentic AI systems and design a 5-layer production blueprint on Azure

2. Deploy and configure Azure OpenAI models, Azure AI Foundry, Cosmos DB, and related services

3. Implement advanced prompting techniques including Chain-of-Thought and ReAct frameworks

4. Build multi-agent workflows using LangChain, Semantic Kernel, and Microsoft AutoGen

5. Integrate long-term memory, custom tool execution, and RAG pipelines using Azure services

6. Deploy Python FastAPI backends and ReactJS frontends to Azure App Service

7. Containerize LLM workloads using Docker and deploy to Azure Kubernetes Service (AKS)

8. Implement event-driven autoscaling with KEDA and secure deployments with Azure Key Vault and Entra Workload ID

9. Design and operate CI/CD pipelines using GitHub Actions and Azure DevOps for GenAI systems

10. Evaluate agentic systems using production metrics and monitor them with Azure Monitor and Grafana

Prerequisites

• Proficiency in Python programming

• Basic familiarity with REST APIs and web development concepts

• Foundational understanding of cloud computing (Azure basics preferred)

• Familiarity with Git and version control workflows

• Basic awareness of machine learning concepts (no deep ML theory required)

Course Outline

Module 1: Foundations of Agentic AI Systems on Azure

1. Understanding how LLMs reason, plan, use tools, and maintain memory to execute complex goals

2. Deep dive into the 5-layer system design: Presentation, Orchestration, Execution, Knowledge, and Infrastructure

3. Provisioning and configuring core Azure services: Azure OpenAI (GPT-4/Omni), Azure Functions, and Cosmos DB

4. Deploying models and navigating the Azure AI Foundry Agent Service interface

Module 2: Building AI Agents and Orchestration Workflows

1. Implementing Role-Based Prompting, Chain-of-Thought (CoT) reasoning, and ReAct (Reason + Act) frameworks

2. Utilizing LangChain, Semantic Kernel (Python/.NET), and Microsoft AutoGen to coordinate agent tasks

3. Designing multi-agent conversation patterns: Two-Agent, Sequential, Group, and Nested chat

4. Building specialized workflow patterns: Prompt Chaining, Routing, Parallelization, and Orchestrator-Workers

Module 3: Memory, Knowledge, and Tool Integration

1. Implementing short-term (ephemeral) and long-term memory using Azure Cosmos DB, MemGPT, or Mem0 for cross-session context

2. Writing and registering custom tools via Azure Functions (REST endpoints) to enable code execution, API calls, and scheduling

3. Building RAG (Retrieval-Augmented Generation) systems using Azure AI Search with semantic and layout-aware chunking and vector embeddings

4. Integrating Microsoft Fabric Data (Lakehouses) and Bing Search into the Foundry Agent Service

Module 4: Web Application Deployment with Azure App Service

1. Leveraging Azure App Service (PaaS) for scalable HTTP-based web apps, mobile backends, and REST APIs

2. Deploying a Python FastAPI backend: configuring Azure Web Apps for Linux, injecting environment variables, and setting startup commands with uvicorn and gunicorn

3. Deploying a ReactJS front-end: setting up Node environments, configuring process managers for SPAs, and connecting to the backend API

Module 5: Scalable Infrastructure and AKS Deployment

1. Writing secure multi-stage Dockerfiles optimized for Python LLM inference workloads

2. Deploying to Azure Kubernetes Service (AKS): provisioning GPU-enabled node pools, managing VNet networking, and writing Kubernetes manifests with zero-downtime health probes

3. Implementing event-driven autoscaling with KEDA to scale GPU-bound agentic workloads based on queue depth, including scale-to-zero cost optimization

4. Implementing passwordless security architectures using Azure Key Vault, Secrets Store CSI Driver, and Microsoft Entra Workload ID

Module 6: CI/CD, MLOps, and Agent Evaluation

1. Automating CI/CD pipelines with Azure DevOps and GitHub Actions: Pull Requests, Docker image packaging, and continuous deployment

2. Implementing safe deployment strategies: Blue/Green and Canary deployments for risk mitigation and instant rollbacks

3. Evaluating agentic systems using production metrics: Task Success Rate, Tool Call Accuracy, Hallucination Rate, and Cost-per-Task with Golden Datasets and LLM-as-a-Judge

4. Monitoring production systems: token cost tracking, distributed tracing with OpenTelemetry, latency analysis, and real-time drift detection using Azure Monitor and Grafana

Module 7: Capstone Project

1. Project Option A — AI Travel Concierge: Build an agent that plans end-to-end trips using external search tools and persistent user memory

2. Project Option B — Agentic Data Analysis Platform: Build a multi-agent system capable of fetching, cleaning, and visualizing datasets

3. Final deliverable: a fully containerized, version-controlled repository deployed to Azure App Service or AKS via automated GitHub Actions CI/CD pipelines