Bloom-aligned objectives 

• Understand: core Azure GenAI building blocks and how they compose into enterprise apps 

• Analyze: reference architectures for chatbots, copilots, and workflow assistants 

• Design: a target architecture with clear boundaries (model, retrieval, tools, app tier, ops) 

Topics 

• Azure AI Foundry concepts: projects, model deployments, app development lifecycle (dev → eval → deploy)  

• Azure OpenAI role in the stack (chat completions, embeddings) and common enterprise app patterns 

• Architecture patterns 

• Chatbots and knowledge assistants (RAG-based) 

• Copilot-style experiences (task-focused, tool-using) 

• Workflow assistants (event-driven, approvals, deterministic steps) 

• Non-functional requirements baseline: latency, availability, auditability, data boundaries 

Labs 

• Lab 1.1: Solution blueprint — Draft a reference architecture and component diagram for a chosen pattern (RAG chatbot / workflow assistant / copilot-like agent). 

• Lab 1.2: Azure resource plan — Define the minimal secure Azure resource set (Foundry/OpenAI, Search, App runtime, Key Vault, Monitor). 

Bloom-aligned objectives 

• Apply: advanced prompting techniques for controllability and reliability 

• Create: prompt flows that orchestrate prompts, tools, and code 

• Evaluate: prompts and flows using test datasets and evaluation runs 

Topics 

• Prompt engineering for production 

• role + task + constraints + format contracts 

• structured outputs (JSON schemas), deterministic templates, refusal behaviors 

• mitigation patterns: ambiguity handling, self-check prompts, “unknown” routing 

• prompt flow in Azure AI Foundry 

• building executable flows that chain prompts, Python tools, and integrations  

• variants, debugging, iteration cycles 

• PromptOps 

• prompt versioning strategy, regression tests, golden datasets 

• Prompt versioning, registry, evaluation and optimization using mlflow 

• evaluation runs and interpreting evaluation metrics in Foundry 

Labs 

• Lab 2.1: Prompt contract pack — Build 6 reusable prompt templates (summarize, extract, classify, rewrite, generate plan, validate). 

• Lab 2.2: Prompt flow pipeline — Implement a multi-step prompt flow: intake → extract entities → generate response → validator step.  

• Lab 2.3: Prompt evaluation suite — Create a test dataset and run evaluations; compare variants and document a promotion decision.  

Bloom-aligned objectives 

• Design: a RAG architecture with clear retrieval and citation strategy 

• Implement: ingestion + chunking + embeddings + indexing in Azure AI Search 

• Analyze: retrieval quality and tune chunking, query strategy, and grounding 

Topics 

• RAG patterns and tradeoffs; why grounding is required for enterprise reliability  

• Azure AI Search for RAG 

• vector and hybrid retrieval concepts 

• index design (metadata, filters, scoring) 

• Content preparation pipeline 

• document parsing and extraction 

• chunking strategies, overlap, semantic boundaries 

• Classic RAG vs agentic retrieval (when retrieval becomes iterative/tool-driven)  

• Implementation blueprint using Microsoft’s RAG tutorial pattern (FastAPI + OpenAI + Search)  

Labs 

• Lab 3.1: RAG ingestion pipeline — Build an ingestion job: parse documents → chunk → embed → index into Azure AI Search. 

• Lab 3.2: Grounded chat API — Implement a FastAPI/Node API that performs retrieval + grounded answer generation with citations. 

• Lab 3.3: Retrieval tuning — Improve answer quality by adjusting chunking/index schema and adding metadata filters; measure improvements with an evaluation set. 

Bloom-aligned objectives 

• Apply: function calling and tool-use patterns for deterministic business workflows 

• Create: orchestrated workflows that combine LLM steps with rules and validations 

• Evaluate: reliability with negative tests and safe-fail behaviors 

Topics 

• “Static workflow” design: deterministic orchestration around an LLM 

• intake validation 

• tool invocation 

• policy checks 

• final response assembly 

• Tool/function calling patterns 

• API tools (CRM lookup, ticket creation, approvals) 

• retrieval tools (search, KB lookup) 

• calculation/transform tools (Python, rules engine) 

• Guardrails in workflows 

• data minimization 

• approval gates (draft vs execute) 

• safe output constraints and redaction 

Labs 

• Lab 4.1: Tool-using assistant — Build an assistant that calls 2 tools (retrieval + business API stub) and returns a structured result. 

• Lab 4.2: Validation-first workflow — Implement a workflow that blocks unsafe/low-confidence outputs and routes to human review. 

Bloom-aligned objectives 

• Understand: agent runtime concepts (threads, runs, messages, state) 

• Implement: single-agent and multi-step agent behaviors with custom tools 

• Design: agent boundaries, memory strategy, and escalation paths 

Topics 

• Foundry Agent Service overview and why agents differ from single-shot chat 

• Conversation state and persistence: threads/runs/messages 

• Tool integration for agents 

• custom tool invocation patterns 

• safe tool execution and permission boundaries 

• Orchestration patterns 

• planner + executor 

• critic/reviewer loop 

• multi-agent handoffs (conceptual alignment to common agent design patterns) 

Labs 

• Lab 5.1: Build an agent with persistent threads — Implement an agent that maintains state across turns and performs a multi-step task. 

• Lab 5.2: Agent with custom tools — Add at least 2 tools (retrieval + action tool) and enforce execution constraints. 

• Lab 5.3: Agentic RAG — Implement iterative retrieval (agent decides when to search again) and compare to classic single-pass RAG. 

Bloom-aligned objectives 

• Implement: production-grade API design for GenAI apps (streaming, retries, timeouts) 

• Deploy: an end-to-end application to Azure runtime services 

• Analyze: performance bottlenecks and apply scaling patterns 

Topics 

• API engineering for GenAI 

• streaming responses, async execution, idempotency 

• concurrency limits and backpressure 

• caching strategies (prompt caching, retrieval caching) 

• Deployment patterns (choose one for labs) 

• Azure App Service / Azure Container Apps / AKS (containerized) 

• Data stores for state and history (where needed) 

• conversation state storage patterns (with clear retention) 

• CI/CD and environment promotion 

• dev/test/prod separation for prompts, indices, and policies 

Labs 

• Lab 6.1: Deploy the RAG API — Deploy your RAG service to App Service or Container Apps with secure configuration. 

• Lab 6.2: Performance drill — Add caching + request controls; document latency and throughput changes. 

Bloom-aligned objectives 

• Apply: Azure OpenAI security baseline principles (identity, network, monitoring) 

• Implement: network isolation (private endpoints) and least-privilege access 

• Evaluate: prompt injection and jailbreak risks; apply mitigations 

Topics 

• Azure OpenAI security baseline: control categories and operational expectations 

• Network security: private endpoints and VNet integration patterns for Azure OpenAI  

• Identity and access 

• managed identities, RBAC, secrets management patterns 

• Safety controls in Azure 

• content filters and policy enforcement in Foundry workflows 

• Prompt Shields (jailbreak/prompt injection defenses) in Azure AI Content Safety 

• Threat modeling for GenAI apps 

• data exfiltration via prompts 

• indirect prompt injection via documents 

• tool abuse and privilege escalation 

Labs 

• Lab 7.1: Secure-by-default deployment — Add private endpoint planning + RBAC roles + secret handling checklist for your app. 

• Lab 7.2: Prompt attack simulation — Run a set of jailbreak/injection test prompts; implement mitigations (Prompt Shields + validation gates). 

Bloom-aligned objectives 

• Monitor: Azure OpenAI and app behavior using Azure Monitor/Log Analytics 

• Evaluate: model/app quality continuously using Foundry evaluation runs 

• Optimize: cost and performance with quotas, throttling strategy, and prompt efficiency 

Topics 

• Monitoring Azure OpenAI with Azure Monitor 

• diagnostics settings, Log Analytics, metrics, and workbooks 

• Foundry evaluation operations 

• running evaluations on models/agents/test datasets and reviewing metrics 

• Cost and performance tuning 

• token efficiency (prompt sizing, summarization, retrieval narrowing) 

• batch vs realtime tradeoffs 

• scaling and throttling strategy under load 

• Operational playbook 

• incident types (quality regressions, retrieval drift, safety events) 

• release management for prompts/flows/indices 

Labs 

• Lab 8.1: Monitoring dashboard — Configure diagnostic settings and create basic Log Analytics queries/workbook tiles for usage and errors. 

• Lab 8.2: Quality regression gate — Set up an evaluation run as a release gate for prompt/flow changes and document go/no-go criteria.