RAG Agents Accelerator

Report Abuse

Basic Information

This repository is an enterprise-focused accelerator and hands-on workshop for building a Retrieval-Augmented Generation (RAG) multi-agent Proof of Concept (3-5 day POC) using Azure services. It provides code, notebooks, deployment scripts and workshop assets to guide teams through designing and implementing a multi-agent architecture that can query diverse data sources, retrieve evidence, and generate contextual answers. The materials target Microsoft delivery teams and customer engineering teams and include backend and frontend components, sample datasets, prerequisites and step-by-step instructions to deploy Azure infrastructure, configure credentials, run notebooks and connect agents to multiple channels for demonstration and evaluation.

Links

App Details

Features
The accelerator is implemented in Python and integrates Azure Cognitive Services for document enrichment and OCR, Azure AI Search with hybrid text and vector search, Azure OpenAI models and text-embedding-3-large embeddings. It uses LangChain for orchestration and LangGraph for a multi-agent architecture. Additional capabilities include multi-index and multi-lingual support, multi-modal input/output (text and audio), tabular Q&A with CSV and SQL sources, API data connectors, SerpAPI for web search, Azure AI Document Intelligence for PDFs, CosmosDB for persistent conversation memory, a Streamlit frontend and backends built with Bot Framework and FastAPI. The README includes model deployment and infrastructure deployment guidance.
Use Cases
The repo accelerates enterprise POC delivery by providing ready-made components, step-by-step workshop modules, and deployment automation so teams can rapidly build, test and demonstrate intelligent agents. It helps technical and non-technical stakeholders evaluate RAG approaches, connect agents to multiple data sources (blob storage, SQL, CSV, APIs, web), and deploy agents across channels like web chat and Teams. The included notebooks, prerequisites checklist and troubleshooting notes reduce setup friction. Persistent memory and multi-channel backends enable realistic demos and evaluation. Provided assets for Microsoft delivery (decks, training guidance and sample questions) support workshop facilitation and customer engagement.

Please fill the required fields*