Project Miyagi provides a hands-on workshop and sample codebase for building enterprise-grade AI-infused applications using Microsoft's Copilot stack. It targets developers and researchers interested in exploring generative AI, prompt engineering, and traditional ML within a cloud-native, event-driven architecture, offering a path to enhanced productivity and personalization.

How It Works

Miyagi showcases a polyglot microservices architecture orchestrated by Semantic Kernel, integrating various LLM frameworks like LangChain and LlamaIndex. It demonstrates techniques such as Retrieval-Augmented Generation (RAG), chain-of-thought prompting, vectorization for memory, and agent-like orchestration. This approach allows for flexible integration of diverse AI models and tools, grounding LLMs with external data and enabling complex, multi-step AI workflows.

Quick Start & Requirements

• Install/Run: Primarily cloud-based deployments (Azure Container Apps, AKS) and local setups via Docker. Specific commands depend on the chosen use case.
• Prerequisites: Azure subscription, Azure OpenAI, Azure AI Search, potentially other Azure services. Python 3.x, Node.js, Docker. Specific models (Llama2, Phi-2) may require AzureML access.
• Resources: Setup complexity varies; cloud deployments require Azure resource configuration.
• Links: intelligentapp.dev, Azure/reddog-solutions

Highlighted Details

• Demonstrates generative text and image use cases (financial coaching, summarization).
• Integrates multiple vector stores (Azure AI Search, CosmosDB pgvector).
• Features fine-tuned models (Llama2, Phi-2) and agent orchestration (AutoGen).
• Includes examples for Semantic Kernel, Promptflow, LangChain, and LlamaIndex.

Maintenance & Community

This is a Microsoft-sponsored sample project. Contributions are welcomed via a CLA. The project follows the Microsoft Open Source Code of Conduct.

Licensing & Compatibility

Provided "as is" for demonstration purposes only. Explicitly stated as not intended for commercial use or any purpose beyond demonstration. No warranties are provided.

Limitations & Caveats

This project is a work in progress with incrementally implemented use cases. The disclaimer strongly advises against relying on this software for any purpose, and it is not intended for commercial use.