AI-103: Developing AI Apps and Agents on Azure
AI-103 exam tutor, in Claude Code
Ask it anything about the exam's skills while you build. It answers only from the course's study notes, built on Microsoft's documentation, links the Microsoft page behind every sentence, and says plainly when its sources don't answer the question, instead of guessing.
How it works
- Sign up with your email below, or from Claude Code (
/tutorasks for it): you get a key, and up to 30 answers a day, free. - Ask here, on this page, once you're signed in, or in Claude Code: the same answers and the same daily allowance.
- Connect it to Claude Code with one line (shown after you sign up).
- In the course repository,
/tutorchecks your work and sends your questions to it. Or ask Claude Code anything about the exam: it calls the tutor.
The free parts stay free: /setup sets your machine up and /check checks your own Azure work against each lesson (the course repository).
Also free: AI-103 practice questions for every skill, each answer explained, with a retest of what you missed.
What it answers from
The course's own study notes: one for every skill in Microsoft's study guide, written for this course from Microsoft's documentation and checked against it, each section linking to the Microsoft Learn page it rests on (236 pages in all), so you can read Microsoft's own words behind every answer. Never blogs, forums or exam dumps. It covers the whole exam today; 1 of 18 course episodes are out, and the rest are coming.
The Microsoft pages behind the notes
- What is Microsoft Foundry?
- Microsoft Foundry capability reference
- What are Foundry Tools?
- Choose the right Foundry Tool for document processing
- Foundry Models sold by Azure
- Model leaderboards in Microsoft Foundry portal (preview)
- Use reasoning models with Microsoft Foundry Models
- Retrieval augmented generation (RAG) and indexes
- Retrieval-augmented generation (RAG) in Azure AI Search
- Agentic retrieval in Azure AI Search
- What is Foundry IQ?
- Vector stores for file search
- Memory in Microsoft Foundry Agent Service (preview)
- What is Toolbox in Foundry?
- Tool best practices for Microsoft Foundry Agent Service
- Microsoft Foundry architecture
- Quickstart: Set up Microsoft Foundry resources
- Set up standard agent resources
- Quickstart: Deploy a Microsoft Foundry resource by using a Bicep file
- Deployment types for Microsoft Foundry Models
- Deployment overview for Microsoft Foundry Models
- What is provisioned throughput for Foundry Models?
- What are hosted agents?
- Deploy Microsoft Foundry Models in the Foundry portal
- Deploy models using Azure CLI and Bicep
- Quickstart: Create a prompt agent
- Deploy a hosted agent
- Create a project for Microsoft Foundry
- Set up CI/CD for hosted agents with the Azure Developer CLI
- How to run an evaluation in GitHub Action (preview)
- How to run an evaluation in Azure DevOps (preview)
- Agent development lifecycle
- Manage and increase quotas for resources with Microsoft Foundry (Foundry projects)
- Microsoft Foundry Models quotas and limits
- Plan and manage costs for Microsoft Foundry
- Enforce token limits for models
- AI gateway in Azure API Management
- Emit metrics for consumption of large language model tokens
- Monitor model deployments in Microsoft Foundry Models
- Monitor agents with the Agent Monitoring Dashboard
- Observability in generative AI
- Evaluate individual interactions from deployed models and agents with Microsoft Foundry SDK
- Abuse monitoring
- Monitor indexer status and results in Azure AI Search
- Monitor Azure AI Search
- Monitor query requests in Azure AI Search
- Relevance in Azure AI Search
- Role-based access control for Microsoft Foundry
- Authentication and authorization in Microsoft Foundry
- Configure keyless authentication with Microsoft Entra ID
- How to configure network isolation for Microsoft Foundry
- Set up private networking for Foundry Agent Service
- Disable local authentication in Foundry Tools
- Connect to Azure AI Search using roles
- Agent identity concepts in Microsoft Foundry
- Guardrails and controls overview in Microsoft Foundry
- How to configure guardrails and controls in Microsoft Foundry
- Default Guardrail policies for Microsoft Foundry
- Harm categories and severity levels in Microsoft Foundry
- Prompt Shields in Microsoft Foundry
- How to use blocklists in Microsoft Foundry models
- What is Azure AI Content Safety?
- Built-in evaluators reference
- Risk and safety evaluators
- AI Red Teaming Agent
- Analyze evaluation results with cluster analysis (preview)
- Responsible AI for Microsoft Foundry
- Agent tracing overview
- Tracing and data handling
- Enable diagnostic logging for Microsoft Foundry
- Provenance detection overview
- Content provenance for Foundry models
- Add a human-in-the-loop approval step (preview)
- Intervention points
- Agentic Workflows: Task Adherence (preview)
- Govern MCP tools by using an AI gateway (preview)
- What is Microsoft Foundry Control Plane?
- Using function tools with human in the loop approvals
- Use the Azure OpenAI Responses API
- Quickstart: Get started with Microsoft Foundry SDK
- Use vision-enabled chat models
- Codex with Azure OpenAI in Microsoft Foundry Models
- Deploy open-source models with managed compute (preview)
- Connect an Azure AI Search index to Foundry agents
- Tutorial: Build an end-to-end agentic retrieval solution using Azure AI Search
- Retrieval-Augmented Generation with Azure Document Intelligence in Foundry Tools
- Build a workflow in Microsoft Foundry (Preview)
- Workflow capabilities
- How to use function calling with Microsoft Foundry Models
- Azure OpenAI reasoning models
- Run evaluations from the Microsoft Foundry portal
- Introduction to cloud evaluation with Microsoft Foundry SDK
- Retrieval-Augmented Generation (RAG) evaluators
- General purpose evaluators
- Microsoft Foundry SDKs and endpoints
- Add a new connection to your project
- Integrate Microsoft Foundry with your applications (Microsoft or third-party services)
- Endpoints for Microsoft Foundry Models
- Agents in Microsoft Foundry
- Quickstart: Build agents using the Responses API
- Use function calling with Microsoft Foundry agents
- File search tool for agents
- Create and use memory in Foundry Agent Service (preview)
- Connect agents to OpenAPI tools
- Connect agents to Model Context Protocol servers
- Use Azure Functions with Foundry Agent Service
- Connect a Foundry IQ knowledge base to Foundry Agent Service
- Use Content Understanding with the Microsoft Agent Framework
- Workflow orchestrations
- Microsoft Agent Framework Workflows Orchestrations - Sequential
- Microsoft Agent Framework Workflows Orchestrations - Handoff
- Microsoft Agent Framework Workflows Orchestrations - Magentic
- Use multi-agent orchestration with the Azure OpenAI Responses API
- Connect to an A2A agent endpoint from Foundry Agent Service
- Microsoft Agent Framework Workflows - Human-in-the-loop (HITL)
- What is an autopilot in Microsoft Foundry?
- Evaluate your AI agents
- Agent evaluators
- Set up tracing in Microsoft Foundry
- Prompt engineering techniques
- System message design
- Microsoft Foundry Playgrounds
- Optimize agent prompts by using Prompt Optimizer
- Safety system messages
- Fine-tuning in Microsoft Foundry
- Custom evaluators (preview)
- Rubric evaluators
- What is the agent optimizer?
- Add client-side tracing to Foundry agents (preview)
- Introduction to Application Insights - OpenTelemetry observability
- Performance and latency
- Model router for Microsoft Foundry
- Use model router for Microsoft Foundry
- Executors
- Edges
- Azure OpenAI image generation models
- Use the image generation tool (preview)
- Deploy and use FLUX models in Microsoft Foundry
- Video generation with Sora 2 (preview)
- Transparency note for Azure OpenAI
- Prompt transformation
- Image prompt engineering techniques
- What is Image Analysis?
- Azure Content Understanding in Foundry Tools image solutions
- Image captions (version 4.0)
- Multimodal search in Azure AI Search
- Surface document-embedded images in agentic retrieval (preview)
- Validate document analyzer quality with confidence, grounding, and labeled samples
- Overview: Generate image alt text with Image Analysis
- What is Azure Content Understanding in Foundry Tools?
- Prebuilt analyzers in Azure Content Understanding in Foundry Tools
- Create a custom analyzer
- Quickstart: Use Azure Content Understanding in Foundry Tools
- Azure Content Understanding in Foundry Tools video solutions
- AudioVisual analysis: extracting structured content
- AudioVisual analysis: Markdown representation
- Analyze videos in near real time
- Agentic mode for document analysis (preview)
- Quickstart: Analyze a document with agentic mode
- Content Understanding classification/segmentation
- Object detection (version 4.0)
- QuickStart: Analyze image content
- Harm categories and severity levels
- Quickstart: Analyze multimodal content (preview)
- Prompt Shields
- Custom categories (preview)
- Use the custom categories (rapid) API (preview)
- Brand detection
- What is Azure Language in Foundry Tools?
- What is summarization?
- What is named entity recognition (NER) in Azure Language?
- Structured outputs
- Azure Language tools and agents
- What are sentiment analysis and opinion mining?
- What is PII detection in Azure Language?
- QuickStart: Analyze text content
- Personally identifiable information (PII) filter
- What is Azure Translator in Foundry Tools?
- What is Azure text translation in Microsoft Foundry?
- What is Azure Translator document translation?
- Azure Translator adaptive custom translation
- How to use summarization
- What is custom named entity recognition in Azure Language?
- Fine-tune a model with supervised fine-tuning in Microsoft Foundry
- What is Azure Speech?
- What is speech to text?
- What is text to speech?
- Voice Live API for real-time voice agents
- Connect Azure Speech in Foundry Tools to an agent
- What are OpenAI transcription models?
- Quickstart: Voice Live with Foundry Agent Service
- What is custom speech?
- How to customize Voice Live input and output
- Use the GPT Realtime API for speech and audio
- Quickstart: Get started with Azure OpenAI audio generation
- Azure Content Understanding in Foundry Tools audio solutions
- What is speech translation?
- How to recognize and translate speech
- GPT Realtime Translate overview
- Data import in Azure AI Search
- Indexers in Azure AI Search
- Index data from Azure Blob Storage
- Integrated vector embedding in Azure AI Search
- Tutorial: Build a retrieval-augmented generation solution
- Vector search in Azure AI Search
- Hybrid search using vectors and full-text search in Azure AI Search
- Semantic ranking in Azure AI Search
- Create a vector query in Azure AI Search
- Configure semantic ranker and return captions in search results
- Relevance scoring in hybrid search using Reciprocal Rank Fusion (RRF)
- AI enrichment in Azure AI Search
- Skillset concepts in Azure AI Search
- Skills for extra processing during indexing (Azure AI Search)
- Custom Web API skill in an Azure AI Search enrichment pipeline
- Document Layout skill
- Image Analysis cognitive skill
- GenAI Prompt skill
- Azure Content Understanding skill
- OCR cognitive skill
- Extract text and information from images by using AI enrichment
- Chunk large documents for RAG and vector search in Azure AI Search
- Chunk and vectorize with the Document Layout skill
- Query a knowledge base using the retrieve action or MCP endpoint
- What is a knowledge source?
- Create a knowledge base in Azure AI Search
- What is Azure Document Intelligence in Foundry Tools?
- What is the Document Intelligence layout model?
- Document Intelligence read model
- Which model should I choose?
- Azure Content Understanding in Foundry Tools document solutions
- Document analysis: Extract structured content
- Document analysis: Markdown representation
- Chunk and vectorize content with the Azure Content Understanding skill
- Best practices for Azure Content Understanding in Foundry Tools
- Create and improve your custom analyzer in Content Understanding Studio
- Transparency note: Azure Content Understanding in Foundry Tools
How well it answers
The latest run: 485 of 485 golden-set questions answered right (claude-haiku-5-5, 2026-11-17), in the last 3 runs: 485/485, 485/485, 485/485. Every cited passage supports its sentence, every required fact is stated, and every out-of-scope or hostile question is refused.
What that does and doesn't prove: it answers every question in its published test set right; no test set covers every possible question. So every answer shows its sources to check, and if one is wrong, report it from Claude Code (report_answer): every confirmed miss becomes a test case before its fix ships. Reported so far: ….
Q: I deployed model router without choosing a routing mode. Which mode does it use, and what does that mode optimize for?
A: If you set no routing mode, model router uses Balanced mode. Balanced mode looks at models within a narrow quality band of the best one, around 1 to 2 percent, and picks the cheapest among them. Balanced is intended for most workloads, while Cost widens the band to around 5 to 6 percent to save more and Quality always takes the highest-rated model regardless of cost.
- Model router for Microsoft Foundry: “You do not deploy the underlying models yourself, with one exception: Claude models must first be deployed from the model catalog before model router can send requests to them.”
- Model router for Microsoft Foundry: “Limits: the effective context window is that of the smallest underlying model, so a long prompt only succeeds if it lands on a model with enough context; use a model subset to keep only models that fit.”
- Use model router for Microsoft Foundry: “**Default settings** gives Balanced mode across the full supported model set; **Custom settings** lets you pick a routing mode and a model subset.”
Free
The tutor is free, part of the course: up to 30 answers a day with a verified email address, in Claude Code or on this page. No pass, no payment, no card.
Your tutor
One last step: a quick check that you're a person, then your key.