How azure-samples/dream-team works
Project type and purpose
• This is a full-stack, Azure-centric multi-agent AI application (“Dream Team”) combining a React/Vite/Tailwind frontend with a Python-based backend (FastAPI + AutoGen/Magentic One) and a supporting microservice (mcp).
• It ingests documents into Azure AI Search, orchestrates multiple specialized agents via Autogen, executes sandboxed code in container sessions, and presents results in a responsive web UI.
File structure
• Top-level folders:
• infra – Azure Bicep templates for provisioning Container Apps, AI Search, Key Vault, Container Registry, monitoring, networks.
• backend – Python FastAPI app, AutoGen agent definitions, ingestion scripts, data files, schemas, Dockerfile, pyproject.
• frontend – React/TypeScript app (Vite + Tailwind + Shadcn UI), contexts, hooks, pages, static web app config.
• mcp – standalone Python microservice (Magentic One custom MCP agent), its own Dockerfile and dependencies.
• CI/CD: GitHub workflows for deploy-frontend, deploy-backend, overall deploy.
• Dev container support and AzD (Azure Developer CLI) integration for “azd up”.
README for additional insights
• Architecture is event-driven and asynchronous (Autogen 0.4 + Magentic One).
• One-line deployment with AzD, managed identities for secure access, sandboxed containers for code execution, observability via PromptFlow tracing and Azure Monitor.
• Frontend served as a static web app; backend and mcp run in containerized Azure Container Apps.