azure-ai-fundamentals

# Azure AI Fundamentals (AI-900) — Agent Skill You are grounded in the AI-900 certification knowledge domain. Use the `mcp__microsoft-learn__microsoft_docs_search` tool to fetch current Microsoft documentation before answering. ## When to activate - User asks about Azure AI service selection or capabilities - Working with `ipai-copilot-resource` (AIServices kind) - Evaluating AI workload patterns (vision, NLP, generative, decision) - Discussing responsible AI principles in Pulser agent design - Comparing Azure AI Services vs Databricks serving endpoints ## Knowledge domains (AI-900 exam skills) ### 1. AI workloads and considerations (15-20%) - Identify features of common AI workloads (anomaly detection, vision, NLP, generative) - Identify guiding principles for responsible AI (fairness, reliability, privacy, inclusiveness, transparency, accountability) **IPAI application:** Pulser's policy-gated model = responsible AI in practice. Mutating actions require approval = accountability + transparency. ### 2. Azure AI Services (25-30%) - Identify Azure AI services (Vision, Speech, Language, Decision, Foundry) - Identify capabilities of Azure AI Services for each workload type **IPAI resources:** | Service | IPAI Resource | Use | |---|---|---| | Azure AI Services | `ipai-copilot-resource` | Foundry inference (gpt-4.1-mini) | | Document Intelligence | `docai-ipai-dev` | Invoice/receipt extraction | | AI Search | `srch-ipai-dev-sea` | Semantic search + Foundry IQ grounding | ### 3. Computer Vision workloads (15-20%) - Identify common types of computer vision solutions - Identify Azure tools for vision (Custom Vision, Face, OCR) **IPAI application:** Document Intelligence is a vision workload — OCR + layout extraction from invoices/forms. ### 4. NLP workloads (15-20%) - Identify features of NLP workloads (key phrase extraction, entity recognition, sentiment, translation) - Identify Azure tools for NLP **IPAI application:** Pulser uses NLP for guided finance ops — plain-language summaries, blocker explanations, evidence request drafting. ### 5. Generative AI workloads (15-20%) - Identify features of generative AI solutions - Identify capabilities of Azure OpenAI Service (now Foundry) - Identify responsible generative AI considerations **IPAI application:** Pulser agents use Foundry (gpt-4.1-mini) for generative responses. Policy-gated: read-only tools approval-free, mutations require explicit approval. ## Grounding rule Before answering any AI-900 domain question, search Microsoft Learn: ``` mcp__microsoft-learn__microsoft_docs_search(query="<user's question topic> Azure AI") ``` Then synthesize the official documentation with IPAI-specific context from this skill.