This workshop demonstrates a prescriptive enterprise approach for deploying Retrieval‑Augmented Generation (RAG) without data duplication. Customers keep their governed, high‑value financial data in Azure NetApp Files as the system of record, expose it through Microsoft Fabric OneLake, and enable Azure AI Foundry agents to retrieve and reason over that data, without migrating or re‑platforming the source datasets.
The architecture establishes a Zero‑Copy AI data path:
- Azure NetApp Files provides performance, security, and enterprise file semantics
- Microsoft Fabric OneLake virtualizes access using native shortcuts
- Azure AI Foundry orchestrates retrieval, grounding, and agent workflows directly against the virtualized data This allows customers to stand up production‑ready RAG experiences faster, reduce data sprawl, and preserve existing governance and operational models.
The outcome is an AI entry scenario for regulated and data‑intensive enterprises: accelerate RAG adoption, minimize architectural disruption, and operationalize AI agents on trusted data—without creating new storage silos or brittle ingestion pipelines.
The solution leverages the Azure NetApp Files object REST API to expose file‑based data through an S3‑compatible object interface, enabling downstream analytics and AI services to access the same data without duplication.
- Storage: Azure NetApp Files (NFS/SMB) with Object REST API enabled
- Data Access: Microsoft Fabric OneLake using S3‑compatible shortcuts
- Indexing & Retrieval: Azure AI Search (OneLake files indexer) for indexing and enrichment, with Azure AI Foundry agents using the indexed content for retrieval‑augmented generation
- Orchestration: Azure AI Foundry agents consuming indexed content for retrieval‑augmented generation
This repository contains all assets required to complete the end‑to‑end workshop and understand the solution architecture being demonstrated.
lab_guide.md: The authoritative, step‑by‑step guide for the workshop. Follow these steps to configure Azure NetApp Files, create a OneLake shortcut, index data with Azure AI Search, and build a grounded AI agent in Azure AI Foundry.walkthrough.md: A narrative and talk track that follows the lab steps, explaining why each step exists and how to position the solution during a live workshop or customer discussion. Use this file to guide presentations and discussions while attendees follow the hands‑on instructions in lab_guide.md.test_data/: Sample unstructured financial data (invoices and CSV logs) used throughout the lab. This data represents realistic enterprise file‑based content and is intentionally stored on Azure NetApp Files to demonstrate AI and analytics operating on file data in place.generate_data.py: A helper script used to generate the sample data included in the test_data directory. This script is provided for transparency and repeatability; it is not required to run the workshop.
Educational Use Only This content is provided for educational and enablement purposes only to illustrate example architectures and “art‑of‑the‑possible” scenarios using Azure services. It is not intended for production use without appropriate review, validation, security hardening, and operational readiness assessments.
The code and documentation are provided as‑is, without warranties of any kind. The authors and Microsoft assume no responsibility or liability for the use of this content in any environment.
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.