What it does Project Miyagi is Microsoft’s own envisioning workshop for its Copilot stack—a polyglot, microservices-heavy sample app meant to show enterprise developers how to weave generative AI into real products. It covers personalized financial coaching, summarization, image generation, and agent-like orchestration across a cloud-native, event-driven backbone.

The interesting bit Rather than picking one framework and calling it a day, Miyagi deliberately piles them all in: Semantic Kernel, LangChain, LlamaIndex, PromptFlow, AutoGen, TypeChat, even a GitHub Copilot Agent VSCode extension. The chaos is the point—it’s a living comparison shop for which Azure AI pieces fit which problems.

Key highlights

• Implements RAG, chain-of-thought prompting, vector memory, fine-tuning (Llama2, Phi-2), and plugin patterns side by side
• Runs on Azure Container Apps or AKS with Cosmos DB, Service Bus, and API Management in the mix
• Includes a working frontend, Jupyter notebook quickstarts, and a ChatGPT plugin sample
• Partial implementations are flagged honestly; the repo is explicitly “Work in Progress”
• Companion workshop materials live at iappwksp.com with a self-guided path

Caveats

• Many listed features are “coming soon” (DeepSpeed Chat/RLHF) or scattered across experiments folders
• The disclaimer repeats four times that this is “for demonstration purposes only” and not production-ready
• Scope creep is real: the tech stack list alone spans 20+ services, which may overwhelm before it educates

Verdict Worth a weekend if you’re an Azure shop architect deciding between Semantic Kernel and LangChain, or need internal buy-in slides with working code behind them. Skip it if you want a clean, minimal starter template—this is the opposite of minimal.