How azure-samples/azure-vector-database-samples works
Main Components
• Code Samples Modules – These are grouped by the specific Azure service used (e.g., Azure AI Search, Azure Cosmos DB for MongoDB vCore, Azure Cosmos DB for PostgreSQL, Azure Database for PostgreSQL, Azure Redis Cache, Fabric Real-Time Analytics (Kusto)). Each of these modules demonstrates similar flows but cater to different underlying vector database technologies.
• Evaluation Samples – Separate components that show how to verify/document performance, query accuracy, or compare data chunking methods.
• Infrastructure-as-Code (IaC) Artifacts – Bicep files and parameter templates that provision Azure resources such as Cognitive Search, PostgreSQL, Redis, Cosmos DB, and the virtual networks (vnet).
• Data Pipelines & Embeddings Ingestion – Jupyter Notebook scripts and environment files (Conda environment YAML files) that orchestrate the process of ingesting and querying embeddings.
• Best Practices – Documentation and guidelines available in the “best_practices” directory which provide supporting standards and directions.
Relationships and Interactions
• Data sources (documents, images, text) feed into pre-processing pipelines (as highlighted by the notebooks) to generate embeddings.
• Infrastructure components (IaC using Bicep) are used to deploy the necessary Azure services – link these resources to their controlling code samples.
• Evaluation samples interact with the deployed infrastructure by running queries or validation steps to measure performance and accuracy.
• Best practices overlay as a guide that informs the design and execution of these code samples and pipelines.
Important Architectural Patterns and Technologies
• Separation of concerns: Different directories isolate the sample implementations while common pipelines and environment templates help unify dependency management across samples.
• IaC pattern: Use of Bicep scripts for automated resource provisioning.
• Data ingestion workflow: Use of Python notebooks and Conda environments to define a reproducible pipeline for generating and querying embeddings.
• Modular design: Each Azure service sample acts as a modular component that can be deployed, tested, and scaled independently.
• External actors: Although not part of the typical running system, note that the repository explicitly does not cover LLM app development, so the focus is solely on vector embedding ingestion and database querying.