Optimizing Supply Chains with Data Streaming and Generative AI

This post details a specific implementation of real-time enterprise RAG within a supply chain context. It describes a system where shop assistants use a GenAI-enabled mobile app to photograph products. This image data is streamed via Confluent to Azure AI's Vision service for scanning and classification. The output is then processed by Apache Flink, joined with real-time inventory data from SQL Server, converted into vector embeddings, and stored in Azure Cosmos DB. This enables retrieval-augmented generation (RAG) by an LLM (OpenAI) to provide actionable instructions to shop assistants, such as 'Keep', 'Remove', or 'Discount', thereby reducing waste and boosting sales. The architecture highlights the integration of computer vision, stream processing, vector databases, and LLMs for real-time decision-making.