Azure for AI/ML Applications#
Introduction to the Azure AI Platform#
Unified Experience: Azure AI Studio vs. Azure Machine Learning studio.
Key Pillars: Foundational Models, Machine Learning, Responsible AI.
Core AI Development & Model Services#
Azure AI Studio#
Hub for generative AI and copilot development.
Prompt Flow for orchestrating LLM workflows.
Evaluation and monitoring tools.
Model Infrastructure#
Azure OpenAI Service: GPT-4o, GPT-4 Turbo, embeddings (text-embedding-3-large), DALL-E 3, moderation, fine-tuning.
Azure AI Model Catalog (via AI Studio): Curated catalog of open-source & frontier models (Llama, Mistral, Phi, etc.) with integrated endpoints.
Custom Model Training: Bring your own model (BYOM) and framework to Azure ML.
Applied AI Services (Azure AI Services)#
Cognitive Services: Vision, Speech, Language, Decision - prebuilt APIs.
Azure AI Search: Hybrid (Keyword + Vector) Search with semantic ranking as the backbone for production RAG systems.
Azure AI Agents: Service for building multi-step, reasoning agents with built-in tools and evaluation.
Azure AI Speech: Text-to-speech avatars, real-time speech-to-speech translation.
Machine Learning Lifecycle (Azure Machine Learning)#
Data & Compute Infrastructure: Notebooks, compute clusters/instances, serverless compute for ML.
Model Training & Experimentation: Automated ML (AutoML), MLflow integration.
MLOps: Model registry, pipelines, endpoint deployment, monitoring.
Core Cloud Infrastructure for AI/ML#
Compute Options#
Virtual Machines: NCasT4_v3, NC A100 v4, ND H100 v5/H200 series for cutting-edge GPUs. Spot instances for cost-effective training.
Serverless Inference: Managed Online Endpoints (with auto-scaling) and Serverless Endpoints (pay-per-execution) in Azure ML.
Kubernetes: Azure Kubernetes Service (AKS) with GPU nodes for scalable inference.
Data & Storage#
Blob Storage & Data Lake Storage Gen2: Primary data store for unstructured data.
Vector Databases: Azure Cache for Redis (with vector search) or integrated vector search in Azure AI Search.
Databases: Cosmos DB (for operational data), Azure SQL Database.
Responsible AI & Governance#
Toolbox for fairness, interpretability, and transparency.
Content safety filters.
Audit trails and compliance (GDPR, etc.).
Architecture Patterns & Best Practices#
Retrieval-Augmented Generation (RAG) with Azure AI Search.
Fine-tuning vs. Prompt Engineering.
Cost Optimization: Spot VMs, managed endpoints, monitoring utilization.
Integration & Ecosystem#
GitHub Copilot & Azure: Integration for developers.
Power Platform: Build AI-powered apps with Copilot Studio.
Microsoft Fabric: Unified analytics platform with AI synergy (OneLake, Synapse Data Engineering, Direct Lake).