SOL-01

Microsoft Foundry — Enterprise AI Agent & Knowledge Platform Azure

Retail Enterprise Client, 45+ locations, 1,200+ employees — modernized to a secured, AI-enabled Azure hybrid estate.

Business Problem

Run 45+ retail stores on trusted AI over a 40,000-SKU catalog, with no vendor lock-in and full governance.

End-to-End Architecture

flowchart LR
  S["Sources
catalog · docs · voice · images · IoT"] --> ING["Ingest
Data Factory · Event Hubs · Durable Functions"]
  ING --> LAKE["Fabric / OneLake
Bronze→Silver→Gold · Databricks"]
  LAKE --> IDX["AI Search
embeddings · hybrid · skillsets"]
  IDX --> FDRY["Microsoft Foundry
Agent Service · Models · MCP/A2A"]
  FDRY --> GUARD["Responsible AI
Content Safety · LLM-as-judge · human-in-loop"]
  GUARD --> GW["Azure OpenAI Gateway
monitoring · custom Entra auth"]
  GW --> APP["Store apps · voice agent · copilots"]
  FDRY --> OBS["Observability
App Insights · Langfuse"]
  GOV["Governance
Agent 365 · Purview · Policy"] -.-> FDRY
  MLO["MLOps
Azure ML · MLflow · Model Registry"] -.-> FDRY

Workflow (Proof of Concept, A→Z)

sequenceDiagram
  participant U as Store user
  participant GW as OpenAI Gateway
  participant AG as Foundry Agent
  participant RS as AI Search (RAG)
  participant GD as Content Safety
  U->>GW: question (auth via Entra)
  GW->>AG: routed prompt + policy
  AG->>RS: retrieve grounded catalog context
  RS-->>AG: top-k passages + citations
  AG->>GD: safety + grounding check
  GD-->>AG: pass / redact
  AG-->>U: grounded, cited answer
  AG->>AG: log trace → App Insights / Langfuse