// the find
Azure-Samples/azure-search-openai-demo
A sample app for the Retrieval-Augmented Generation pattern running in Azure, using Azure AI Search for retrieval and Azure OpenAI large language models to power ChatGPT-style and Q&A experiences.
Microsoft's reference implementation for RAG on Azure: upload your docs, deploy with one command, get a ChatGPT-style Q&A interface backed by Azure AI Search and Azure OpenAI. It's a starting point, not a production system — the README says this explicitly. Aimed at enterprise teams who are already Azure-committed and want a working skeleton to fork.
The azd-based deployment is genuinely good — one command provisions everything including search indexes, blob storage, and the app itself. The document ingestion pipeline handles PDFs, images, JSON, and CSV with real chunking logic, not naive line splits. Citation rendering with thought-process transparency is built in and works well for demonstrating retrieval quality to stakeholders. The eval harness in /evals is a real differentiator — ground truth files, scoring scripts, and saved results across multiple model configs let you actually measure retrieval quality instead of eyeballing it.
The security posture is a known problem the repo itself flags — RBAC, network isolation, and auth are all opt-in and off by default, so forks that skip the productionizing guide will deploy publicly open endpoints. The chunking strategy is fixed-size with overlap and doesn't adapt to document structure, which means tables and code blocks get split mid-row and retrieval quality drops for technical docs. Local dev requires a full Azure deployment first — there's no offline mode or local mock for the search backend, so iteration is slow and costs money. The TypeScript frontend is a single monolithic Chat.tsx page; the moment you want multi-index or multi-tenant routing you're doing a major rewrite.