The developer-focused Microsoft Learn course behind the Azure AI Apps and Agents Developer Associate certification (AI-103). Across four learning paths it covers building generative AI apps, AI agents, natural language and speech solutions, and visual-data insight solutions — all on Microsoft Foundry.
Each chapter here distils one module — its units condensed into the key concepts, paired with a quiz. These are independent study notes; take the free, hands-on modules on Microsoft Learn for the full course.
Original work & credit
Microsoft
Free, self-paced Microsoft Learn course (AI-103) for the Azure AI Apps and Agents Developer Associate certification.
- Title
- Develop AI Apps and Agents on Azure
- Publisher
- Microsoft Learn
- Source
- Microsoft Learn course AI-103T00
These notes are independent, original summaries created for study purposes. All rights to the underlying work belong to Microsoft and Microsoft Learn. Please support the authors by taking the original training.
Chapters
30 of manyPlan and Prepare to Develop AI Solutions on Azure
The five developer AI capabilities, Microsoft Foundry projects and their four asset types (models, agents, tools, knowledge), the Foundry Tools catalog and SDKs, and the six responsible-AI principles.
Select, Deploy, and Evaluate Microsoft Foundry Models
The Foundry Models catalog and model types, selecting with benchmarks across quality/safety/cost/performance, the nine deployment types, and manual + automated model evaluation.
Develop a Generative AI Chat App with Microsoft Foundry
Choosing an endpoint and SDK (Foundry SDK vs OpenAI SDK), the stateful Responses API with previous_response_id, and the ChatCompletions API's system/user/assistant roles.
Develop Generative AI Apps That Use Tools
Tools and the tool-calling loop, the four built-in tools (code_interpreter, web_search, file_search, function), and the function-calling loop where the model requests and your code executes.
Optimize Generative AI Model Performance with Microsoft Foundry
The optimization spectrum — prompt engineering (patterns, parameters), RAG grounding with embeddings and Azure AI Search, fine-tuning (LoRA, SFT/RFT/DPO), and how to compare and combine the three.
Implement a Responsible Generative AI Solution in Microsoft Foundry
The four-stage process (map, measure, mitigate, manage), mapping and prioritizing harms with red-teaming, and the four mitigation layers including content filters (four severity levels, five harm categories).
Develop AI Agents with Microsoft Foundry and Visual Studio Code
What AI agents are and the fully-managed Foundry Agent Service, declarative vs hosted agent types and their components, the portal-vs-VS-Code development approaches, and agent security risks and mitigations.
Integrate Custom Tools Into Your Agent
The four custom-tool options (custom function, Azure Functions, OpenAPI spec, Logic Apps) and how to integrate them via FunctionTool/AzureFunctionTool/OpenApiTool under the declarative pattern.
Integrate MCP Tools with Azure AI Agents
The Model Context Protocol and dynamic tool discovery, the server/client/agent architecture, and connecting an Azure AI agent to a remote MCP server via the MCPTool with its approval flow.
Build Knowledge-Enhanced AI Agents with Foundry IQ
Foundry IQ as managed, shared retrieval — domain knowledge bases, automatic data-source indexing and built-in retrieval intelligence, the MCP connection, and configuring instructions for grounded, cited answers.
Integrate Your Agent with Microsoft 365
Publishing options for M365 and Teams, the Agent Application resource and its shared vs organization scopes, and Work IQ for secure, permission-aware access to Microsoft 365 data.
Build Agent-Driven Workflows Using Microsoft Foundry
Visual, declarative workflows — nodes, edges, and managed state; the sequential, human-in-the-loop, and group-chat patterns; and Power Fx logic with human oversight built in.
Develop an AI Agent with the Microsoft Agent Framework
The next generation of Semantic Kernel and AutoGen — one Agent base class, seven core components, provider flexibility with Foundry as the recommended provider, and adding tools with the @tool decorator.
Orchestrate a Multi-Agent Solution Using the Microsoft Agent Framework
Workflow building blocks (executors, edges, events) and the five orchestration patterns — concurrent, sequential, handoff, group chat, and Magentic — all reached through one unified interface.
Discover Azure AI Agents with A2A
The A2A protocol for cross-vendor agent collaboration — the Agent Card and skills, the agent executor and server, and how a client connects to consume a remote agent.
Analyze Text with Azure Language in Foundry Tools
The pre-trained text service — provisioning and authenticating, language detection with confidence scores, named entity recognition across categories, and PII detection with redaction.
Develop a Text Analysis Agent with the Azure Language MCP Server
The Azure Language MCP server and its four tools, autonomous runtime tool selection, and connecting it from the Foundry portal or in code with an MCPTool.
Develop a Speech-Capable Generative AI Application
Speech through generative models — choosing a gpt-4o speech model, transcribing audio to text, and synthesizing text to a voice through the AzureOpenAI client.
Create Speech-Enabled Apps with Azure Speech in Foundry Tools
The purpose-built Speech SDK — SpeechConfig, the shared four-step STT/TTS pattern, audio formats and neural voices, and SSML for fine control of synthesized output.
Develop a Speech Agent with the Azure Speech MCP Server
The Azure Speech MCP server's recognize and synthesize tools, why it requires blob storage and a SAS URL, and connecting it from the portal or code.
Develop an Azure Speech Voice Live Agent in Microsoft Foundry
Real-time speech-to-speech — the Voice Live API's event-driven WebSocket model, the async Python SDK with interruption handling, and building an agent that carries its own configuration.
Translate Text and Speech with Microsoft Foundry Tools
Dedicated translation — why specialized services beat general LLMs, Azure Translator for text and documents, and Azure Speech translation to text or synthesized speech.
Develop a Vision-Enabled Generative AI Application
Chat that can see — multimodal models in Foundry and building a vision chat app with multi-part prompts (URL or Base64) through the Responses or ChatCompletions API.
Generate Images with AI
Text-to-image — how generation differs from search, the gpt-image-1 and FLUX models, the Images playground, and a client app that generates and saves an image.
Generate Videos with Microsoft Foundry
Text-to-video with Sora 2 — deploying the model, prompt-writing craft, and the asynchronous create/poll/download job pattern with reference images and remixing.
Analyze Images with Content Understanding
Schema-driven image analysis — the Content Understanding analyzer framework, prebuilt analyzers and the extract/classify/generate field schema, and reading results with confidence and grounding.
Create a Multimodal Analysis Solution with Azure Content Understanding
Content Understanding end to end — the four content modalities, the schema-to-analyzer build process in Studio, and the asynchronous submit/poll API workflow.
Create an Azure Content Understanding Client Application
Content Understanding in code — setup and required model deployments, creating an analyzer from a JSON schema via SDK or REST, and analyzing content to read typed fields with confidence.
Extract Data with Azure Document Intelligence
Structured extraction from documents — the read/layout analysis models, prebuilt models for common document types, and custom template, neural, classifier, and composed models.
Create a Knowledge Mining Solution with Azure AI Search
The course capstone — Azure AI Search's indexer enrichment pipeline, built-in and custom AI skills, Lucene querying, and persisting projections in a knowledge store.