Report Abuse

Basic Information

Project Miyagi is an envisioning sample and workshop codebase that demonstrates how to design, develop, and deploy enterprise-grade intelligent applications using Microsoft’s Copilot Stack. It is built as a polyglot collection of microservices and experiments that showcase generative and traditional machine learning use cases, agent-like orchestration, and cloud-native event-driven architectures. The repository is intended as a hands-on learning and reference resource for teams exploring prompt engineering, retrieval-augmented generation, vectorized long-term memory, fine-tuning of foundation models, and integration patterns for agents and plugins. The content includes notebooks, demos, sample services, and architecture diagrams to guide implementation and experimentation rather than a single production product.

Links

Categorization

App Details

Features
The repo contains multiple implemented experiments and example artifacts including an MVP with personalization and chat on container apps, notebooks for retrieval-augmented generation, agents using Assistants API and Autogen, a VSCode extension example for a GitHub Copilot Agent, a ChatGPT plugin sample, and knowledge graph memory experiments. It demonstrates use of Semantic Kernel, Promptflow, LangChain, LlamaIndex, various vector stores and embeddings integrations such as Qdrant and CosmosDB/pgvector, and generative image utilities including DreamFusion and ControlNet. The codebase emphasizes an event-driven microservices backbone, architecture diagrams, front-end samples, and references to fine-tuning workflows and AzureML model artifacts.
Use Cases
Miyagi helps developers, architects, and product teams learn practical patterns and end-to-end flows for building AI-infused applications and copilots. It provides hands-on examples to explore prompt engineering techniques like chain-of-thought and in-context learning, methods for retrieval augmentation and vector memory, and approaches to fine-tuning and deploying pre-trained foundation models. The repository offers quickstarts, sample notebooks, and service breakdowns to accelerate prototyping and modernization of applications, guidance on cloud-native operational concerns such as scalability and availability, and reference integrations with common Azure services to move from experimentation toward production-ready designs.

Please fill the required fields*