Microsoft Certified: Azure AI Apps and Agents Developer

2.1 Build generative AI applications with Foundry

• · Deploy and consume LLMs, small language models, code models, and multimodal models
• · Implement retrieval-augmented generation (RAG) against grounded knowledge
• · Design tool-augmented and multistep reasoning workflows
• · Evaluate models for fabrications, relevance, quality, and safety
• · Integrate via Foundry SDKs and connectors; configure app-to-Foundry-project connections

2.2 Build agents with Foundry Agent Service

• · Define agent roles, goals, conversation tracking, and tool schemas
• · Integrate retrieval, function calling, and memory into an agent
• · Integrate agent tools — APIs, knowledge stores, Azure AI Search, Content Understanding, custom functions
• · Implement orchestrated multi-agent solutions (router/planner, sequential handoff, parallel/concurrent)
• · Build autonomous and semiautonomous workflows with safeguards and approval flows

2.3 Optimize and operationalize generative AI

• · Apply prompt engineering and parameter tuning (temperature, top-p, max tokens)
• · Implement model reflection, chain-of-thought, and self-critique loops
• · Add observability — tracing, token analytics, safety signals, latency breakdowns
• · Orchestrate multiple models and hybrid LLM-plus-rules-engine designs
• · Integrate agent monitoring, behavior evaluation, and error analysis

3.1 Generate and edit visual content

• · Generate images and video from text and reference media
• · Edit images — inpainting (fill masked regions), mask-based targeted edits, prompt-driven modifications
• · Implement video editing workflows and generation/editing controls

3.2 Implement multimodal understanding

• · Analyze visual context and caption single or multiple images
• · Implement visual question answering over images
• · Generate accessibility alt-text (concise) and extended descriptions
• · Extract visual characteristics with Azure Content Understanding (single-task vs pro-mode pipelines)
• · Run video analysis workflows and object/region identification

3.3 Apply responsible AI for multimodal content

• · Filter unsafe and disallowed visual content
• · Defend against indirect prompt injection via text embedded in images
• · Enforce visual policy rules — watermarks, prohibited symbols, brand usage, inappropriate content detection

4.1 Analyze text with language models and Foundry Tools

• · Extract entities, topics, summaries, and structured JSON via generative prompting and Foundry Tools
• · Detect sentiment, tone, safety, and sensitive content as distinct classification tasks
• · Translate with Azure Translator in Foundry Tools or with LLM-powered translation flows
• · Customize for domains (e.g., compliance summarization)

4.2 Implement speech solutions

• · Implement speech-to-text and text-to-speech for agentic interactions
• · Use speech as a first-class agent modality, including custom speech models for domain jargon
• · Perform multimodal reasoning from audio input
• · Implement speech translation via language models or Foundry Tools

5.1 Build retrieval and grounding pipelines

• · Ingest and index documents, images, audio, and video for grounding
• · Implement semantic, vector, and hybrid search with Azure AI Search
• · Enrich indexing with built-in skills (native) and custom skills (hosted function/API)
• · Design RAG ingestion flows including OCR, chunking, embeddings, and grounding metadata
• · Connect retrieval pipelines to workflows and agent tools

5.2 Extract information from documents

• · Combine OCR, layout analysis, and field extraction as distinct pipeline stages
• · Build multimodal extraction pipelines with Azure AI Document Intelligence
• · Produce grounded representations via Content Understanding for agents and RAG
• · Configure analyzers for structured and markdown outputs