Azure AI Foundry Application Development Training Course

Course overview

Azure AI Foundry gives development teams a managed workspace for building generative AI applications, but moving from a prompt demonstration to a dependable internal or customer-facing service requires more than selecting a model. Teams must choose an appropriate model and deployment, ground responses in governed enterprise content, control tool access, evaluate quality systematically, trace failures, and protect credentials and sensitive data. This course addresses the practical engineering decisions that determine whether an AI application can be supported in production.

Participants build applications in Azure AI Foundry using projects, model deployments, agents, prompt assets, evaluations and observability. They learn to design retrieval-augmented generation (RAG) with Azure AI Search, define instructions and tool contracts for agents, implement responsible AI controls, compare model quality against test datasets, and diagnose latency, token and failure patterns through traces. The course also covers identity, secrets, role-based access control, deployment pipelines and release criteria for AI workloads.

Instruction combines focused technical demonstrations with guided build sessions in a structured Azure sandbox. Each participant develops a production-oriented application blueprint: an AI Foundry project design, grounded application flow, evaluation dataset and rubric, security model, telemetry plan and release checklist. This package can be adapted for a live product backlog, proof of concept or internal automation initiative immediately after the course.

The course is best suited to developers, architects and technical leads who already work with Azure or APIs and need to deliver governed generative AI applications rather than isolated chatbot prototypes.

Course objectives

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

• Configure an Azure AI Foundry project with model deployments, connections and role-based access controls
• Design a retrieval-augmented generation pipeline using Azure AI Search indexes, chunking and citation prompts
• Build an AI Foundry agent with system instructions, tool definitions and controlled conversation state
• Select and compare foundation models using quality, latency, context-window and cost criteria
• Create evaluation datasets, grading rubrics and automated evaluators for groundedness, relevance and safety
• Instrument an AI application with traces to diagnose model calls, retrieval results, tool execution and token use
• Implement managed identity, Azure Key Vault secrets and content-safety controls for an AI application
• Produce a production release plan containing architecture decisions, test evidence, monitoring thresholds and rollback actions

Benefits of attending

Target competencies

Who should attend

• AI Application Developers — who need to build and ship Azure-hosted generative AI services
• Cloud Solution Architects — who define secure, scalable application patterns for AI workloads
• Software Engineers — who integrate language models, retrieval and tools into existing products
• Technical Leads — who must set engineering standards for AI proof-of-concepts moving into production
• Data Engineers — who prepare searchable enterprise content for retrieval-augmented applications
• DevOps and Platform Engineers — who operationalise identity, deployment, monitoring and governance for AI services

Requirements and prerequisites

Participants should be comfortable writing or reviewing application code in Python, C# or JavaScript and calling REST APIs or SDKs. They should understand core Azure concepts including subscriptions, resource groups, Microsoft Entra ID, role-based access control and basic networking. Prior exposure to JSON, Git and command-line tooling is assumed. Familiarity with language-model prompting and vector search is useful but not essential; these are taught from an application-engineering perspective. Participants need access to an Azure subscription or the course sandbox. Prior data science, model training, fine-tuning or advanced machine-learning mathematics is not required.

Training methodology

The instructor introduces each Azure AI Foundry capability through an architecture decision and live build, then participants apply it in a sandbox project. Labs progress from project setup and model deployment to grounded retrieval, agent tools, evaluation runs and production telemetry. Small groups review a realistic enterprise assistant case, challenge its security and quality assumptions, and agree release gates. On the final day, participants turn their working build into an application plan with named owners, evidence requirements, monitoring measures and next implementation steps.

Course outline

Day 1: Azure AI Foundry foundations and application architecture

• Azure AI Foundry hubs, projects and resource relationships
• Model catalog selection and deployment options
• Azure OpenAI deployment configuration and quota considerations
• Foundry project connections and endpoint management
• Microsoft Entra ID authentication and Azure RBAC roles
• Application architecture patterns for chat, RAG and agent solutions
• Prompt asset design, system messages and parameter controls

Workshop: Configure an Azure AI Foundry project and produce an architecture decision record for a proposed internal knowledge assistant.

Day 2: Grounded applications with retrieval-augmented generation

• RAG failure modes: hallucination, stale content and unsupported claims
• Document ingestion, parsing and metadata design for Azure AI Search
• Chunking strategies, overlap settings and semantic segmentation
• Embedding models and vector, keyword and hybrid retrieval
• Azure AI Search index schemas, filters and relevance tuning
• Prompt patterns for citations, abstention and source attribution
• Retrieval quality testing with representative user questions

Workshop: Build a hybrid Azure AI Search index and deliver a cited RAG response flow for a supplied policy-document corpus.

Day 3: Agents, tools and controlled application behaviour

• Azure AI Foundry Agent Service concepts and agent lifecycle
• System instructions, conversation state and thread management
• Function tool schemas and reliable tool-call contracts
• Grounding tools and enterprise data access boundaries
• Multi-step task design and human approval checkpoints
• Prompt injection threats and indirect-instruction defences
• Agent failure handling, retries and user-facing fallback responses

Workshop: Implement an agent that retrieves approved knowledge and calls a mock service tool while enforcing an approval step for sensitive actions.

Day 4: Evaluation, safety and observability

• Evaluation datasets, expected outputs and adversarial test cases
• Quality metrics for relevance, groundedness, coherence and completeness
• Automated evaluators and comparative model experiments
• Azure AI Content Safety categories and configurable thresholds
• Red-teaming prompts for jailbreak, data leakage and unsafe output
• Application tracing for model calls, retrieval and tool execution
• Token, latency and error analysis for operational decisions

Workshop: Run an evaluation suite against two application configurations and produce a quality, safety and performance recommendation.

Day 5: Production delivery and operational governance

• Managed identities, Azure Key Vault and secret rotation
• Private endpoints, network isolation and data-residency considerations
• CI/CD patterns for prompts, application code and infrastructure
• GitHub Actions deployment gates and environment promotion
• Model versioning, change control and regression testing
• Service-level objectives, alert thresholds and incident triage
• Production readiness reviews and AI application release checklists

Workshop: Create and present a production release pack containing the solution architecture, evaluation evidence, security controls, telemetry plan and phased rollout checklist.

Tools & standards covered

Azure AI Foundry, Azure AI Search, Azure Key Vault, GitHub Actions

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions