AI-103: Azure AI Apps & Agents Developer Exam Prep

Discover how to navigate the course structure for AI-102 and AI-103, identify sections aligned with each exam, and tailor your study plan before the AI-102 retirement.

Learn the fundamentals of artificial intelligence and machine learning, and distinguish predictive ai from generative ai with azure-based use cases and large language model concepts.

Explore AI agents and compound AI systems, where an LLM uses APIs and external tools to autonomously plan and execute workflows, boosting ROI and automating business processes.

Explore core generative AI jargons like tokens, system prompts, user prompts, chat completions API, and multimodal versus unimodal models, highlighting how tokens drive costs and model behavior.

Discover Microsoft Foundry, an ai ecosystem platform as a service for building agentic ai at scale. Includes model catalog, agent service, Foundry IQ, and governance.

Deploy a GPT 5.2 chart model from the Microsoft Foundry model catalog to a serverless API, then explore guardrails, token per minute rate limits, and memory in the playground.

Access the public GitHub repository AI102 Certification for all labs, including language service, speech service, AI vision, Microsoft Foundry, and information extraction solutions.

Explore how to call a deployed OpenAI model using the Microsoft Foundry OpenAI SDK and OpenAI SDK, configuring environment variables, az login, and a Jupyter notebook.

Deploy open vendor-agnostic models in Microsoft Foundry and call them with the Anthropic SDK. Configure environment variables, endpoints, and API keys, then translate English to Spanish using a system prompt.

Explore prompt engineering by learning to craft precise prompts for AI agents and generative applications, focusing on four parts: goal, context, expectations, and source, to maximize accurate, meaningful outputs.

Explore chain of thoughts prompting, few short prompting, and zero short prompting, along with best practices for shaping LLM behavior and output formats.

Discover the Microsoft Foundry SDK, a comprehensive Azure toolchain that unifies models, data, and AI services to build agents with multi-model support and plugin integrations.

Create a code interpreter enabled Microsoft Foundry agent that analyzes an uploaded CSV to generate and download a column chart image, using the Foundry SDK in a code-first workflow.

Explore MCP servers and the model context protocol, enabling AI agents to call APIs via a secure, scalable server without altering agent code.

Learn to build an MCP-enabled Foundry agent that can search Microsoft Learn content and code samples, grounding responses in Microsoft docs and Azure AI resources.

Create a multi-tool Foundry agent using a code-first approach, wiring a weather open API tool and a Microsoft Learn MCP server tool to answer weather queries and find learning modules.

Explore red teaming for AI agents with Microsoft Foundry, PyRIT, and the SDK. Learn how to automate adversarial probing, assess risks, and integrate testing in development and testing phases.

Explore red-teaming to evaluate an AI agent with groundedness, retrieval, coherence evaluators, plus tool call accuracy. Build evaluations, generate attack queries for risk categories, and apply jailbreak and base64 strategies.

Explore low-code, multi-agent orchestration workflows in Microsoft Foundry that empower citizen developers to build scalable AI pipelines, with a human-in-the-loop and agent chat workflows.

Explore the GitHub repository for Microsoft Foundry on my public profile to access hands-on demos and labs in multi-agent workflows.

Build a sequential orchestration workflow in the Foundry ecosystem using three agents—topic builder, MS Learn module picker, and study plan generator—to create a week-by-week azure ai engineering learning path.

Build a human-in-the-loop sequential workflow for the study plan generator in this hands-on lab, adding user feedback with an ask-a-question step and routing back to the topics builder agent.

Explore the Microsoft Agent Framework as a stable, enterprise-ready platform for building multi-agent workflows in the open agentic web, with memory, API calls, and cross-framework interoperability.

Create a simple Batman agent in a Foundry project using the Microsoft Agent Framework, wiring endpoint and deployment name, then explore multimodal, streaming and tracing capabilities.

Visualize the sequential workflow built with Microsoft Agent Framework using devui, connecting researcher and writer agents into a two-node workflow with a single edge and live tracing on port 8090.

Build a parallel workflow with the Microsoft Agent Framework and Foundry Agents to run location picker, destination recommender, weather, and cuisine suggestion, then aggregate results with an itinerary planner.

Discover how the agent-to-agent communication protocol enables interoperable, agent-native conversations between client and remote agents, with MCP servers connecting tools in the Azure AI ecosystem.

Explore the architectural anatomy of the A2A protocol, detailing client and remote agents, OpenID and OAuth, and the JSON RPC data stream. Explain task IDs, context IDs, artifacts.

Explore the Microsoft Foundry GitHub repository for the A2A labs, featuring Python notebooks and a dedicated A2A folder to guide hands-on labs, demos, and activities for agent-to-agent communication.

Expose a foundry agent via an a2a server and connect a client to route queries through the server. Learn to set up environment variables, endpoints, and a Batman agent in Foundry.

Explore how to combine the A2A protocol with an MCP server to create and expose a Foundry agent, enabling streaming responses and access to Microsoft Learn documentation.

Publish your Foundry agent to the Microsoft 365 ecosystem via the Foundry portal or the 365 agents toolkit, enabling Teams and Copilot. Explore two publishing paths and distribution scopes.

Publish a Foundry agent to the Microsoft 365 ecosystem from the Foundry portal, attaching the MCP server. Then access it in Teams and Copilot via the M365 agent store.

Perform a hands-on lab to implement agent tracing with the OpenTelemetry SDK and an Application Insights resource, verify Log Analytics workspace, and view traces in the Microsoft Foundry portal.

Learn to run kql queries in the log analytics workspace, inspecting the traces table and custom dimensions to map user queries to assistant responses using Azure OpenAI and Foundry SDK.

Fine-tune pretrained language models with domain task pairs to tailor outputs and brand voice for business use cases like support templates and policy classification.

Explore sample supervised fine-tuning and direct preference optimization datasets from the AI300 GitHub repository, including SFT and DPO JSONL files with system, user, and assistant prompts.

Explore a hands-on walkthrough of fine-tuning a foundational large language model in Microsoft Foundry, covering model selection, customization methods, data sources, validation, deployment options, and cost considerations.