Skip to content

reference_architecture.md

Latest commit

 

History

History
39 lines (25 loc) · 2.92 KB

reference_architecture.md

File metadata and controls

39 lines (25 loc) · 2.92 KB

Reference Architecture

This section presents a reference architecture for GenAIOps. It highlights how Azure AI Foundry, AI Agent Service and Container Apps can be combined to support data ingestion, experimentation, CI/CD and scalable deployment. Use it to compare against your own designs, adapt key patterns to your environment and accelerate decision-making around infrastructure, governance and operational workflows.

Header

  1. Data estate
    All raw and curated datasets are securely stored and managed in Azure Storage, Data Lake Storage, Azure SQL Database and Azure Cosmos DB, forming a single source of truth that supports grounding, indexing, model development and evaluation.

  2. Administration & setup
    Azure AI Foundry project resources—including models, code repositories, datasets, connections and role-based access—are provisioned and configured through IaC templates to enforce consistency and governance.

  3. Experimenting, Building and Augmenting (inner loop)
    Data scientists iterate rapidly on model selection, prototyping and fine-tuning within the Azure AI Foundry dev environment, using CI-backed container images to validate performance.

  4. Continuous Integration Pipeline
    Every change to code, prompts or model artifacts is automatically tested, containerized and pushed to Azure Container Registry to guarantee reproducible builds.

  5. Operationalizing (outer loop)
    Container images are deployed to Azure Container Apps and AI Agent Service environments via continuous deployment pipelines, promoting rapid rollout across environments.

  6. Development environment
    Within the DEV AI Foundry project, agents are orchestrated and deployed as long-running Container Apps, with manual and batch inference tests ensuring baseline quality.

  7. Staging/test environment
    The QA AI Foundry project hosts A/B and safety evaluations, integration tests and batch inference workloads in Container Apps to validate release candidates under production-like conditions.

  8. Production environment
    Approved AI agents and containerized services run in the PROD AI Foundry project on Azure Container Apps and Kubernetes via Azure Arc, with blue/green deployments, smoke tests and managed batch endpoints ensuring reliability.

  9. AI service monitoring
    Azure Monitor tracks AI Agent Service metrics—such as flow executions, user interactions and model performance—to maintain real-time visibility into service quality.

  10. Infrastructure monitoring
    Azure Application Insights captures container-level telemetry—including availability, latency and resource utilization—for all Container Apps and backend services.

  11. Alerting & notifications
    Scheduled and threshold-based alerts are configured in Azure Monitor to trigger notifications for pipeline failures, performance regressions or capacity issues, enabling proactive operations.