The underlying systems that connect generative AI models to external documents or databases for grounded responses. It helps organizations improve answer relevance, reduce hallucinations, and use proprietary knowledge in AI applications.
Use cases
6
Examples
6
Industries
5
Timeline
5 mo
Data updated 1 day ago
Adoption over time
Documented cases per month
By case publish month · completed months only
6 cases documented across 24 months (Aug 24 – Jul 26), peaking at 2 in May 2025.
AI Use Cases Hub
Each column counts every documented case of this type by its publish month, across the full corpus. The in-progress current month is excluded from columns and surfaced separately, and cases published before the charted window are summarized as earlier cases instead of plotted.
4Innovativeness4/5Advanced4/5 - Advanced. The article describes an advanced multimodal RAG system that combines vector search, graph knowledge, and Amazon Bedrock Data Automation to operationalize maintenance knowledge across languages and media types.
FAW Group Import & Export Co., Ltd. operates Hongqi Overseas’ remote maintenance support system for global vehicle service. The team needed to turn large volumes of maintenance manuals, work orders, images, and video into reusable knowledge assets so dealers and technical experts could resolve issues faster across regions.Using AWS generative AI services, Hongqi built a multimodal RAG-based knowledge system to support natural-language Q&A, cross-language understanding, source-attributed retrieval, and automated maintenance report generation for after-repair knowledge accumulation.
3.7Innovativeness3.7/5Advanced3.7/5 - Advanced. More advanced than standard virtual-agent cases because it combines centralized multi-locale RAG, metadata-driven regional filtering, versioned knowledge bases, automated evaluation/rollback, and serverless orchestration; compared with recent incremental customer-support copilots, this shows a stronger operating model and architecture but not a breakthrough novel pattern.
Ring, Amazon’s home security subsidiary, built a production-ready multi-locale RAG support chatbot using Amazon Bedrock Knowledge Bases.The solution addresses the need for region-specific support content across 10 international regions, beyond simple translation, while keeping latency and infrastructure costs under control.
3Innovativeness3/5Differentiated3/5 - Differentiated. It describes a custom RAG engine (TasteMaker) that generates and version-controls design assets on demand with strict brand control, accelerating timelines from weeks to hours.
Kraft Heinz modernized its internal design processes by leveraging generative AI to scale content creation and personalization while maintaining brand consistency and data security.
4Innovativeness4/5Advanced4/5 - Advanced. UBS built a governed, multi-tenant AI platform combining hybrid RAG (vector+metadata+relational), document intelligence ingestion, MLflow evaluation, and agentic/graph-based retrieval orchestrated on Kubernetes/PostgreSQL, with large-scale self-service provisioning improvements.
UBS, a global investment bank and asset manager, scaled advanced AI for financial services by building the RiskLab AI Common Ecosystem (AICE) on Azure. Supporting over 1,200 data scientists, AICE offers governed generative AI, RAG, and agentic capabilities to accelerate risk, compliance, and investment analysis. Central services include managed LLM endpoints, MLflow-based evaluation, tenanted environments, and robust data governance—all orchestrated via Azure Kubernetes Service and Azure Database for PostgreSQL.AICE's VEGA platform provides governed, self-service lifecycle management for vector stores powering hybrid and hierarchical RAG techniques. UBS uses advanced search—combining metadata, vector, and relational queries—to interrogate thousands of financial documents, supported by ChatGPT, Azure OpenAI, and Azure AI Document Intelligence for layout recognition. The architecture enables rapid provisioning of resources (from days to seconds) and reduces the time to production for new models from six months to one. UBS plans to extend GraphRAG features for macroeconomic scenario analysis and investment forecasting, enabling sophisticated graph-based queries handled by AI agents.
4Innovativeness4/5Advanced4/5 - Advanced. It details a plug-and-play RAG connector using Semantic Kernel with embeddings generation, Elasticsearch vector retrieval, and prompt orchestration designed for modular swapping of components.
Elastic, a global search company, partnered with Microsoft to launch the Semantic Kernel Elasticsearch Vector Store Connector, enabling seamless integration between Microsoft Semantic Kernel, Azure OpenAI, and Elasticsearch for building enterprise AI agents.The collaboration specifically targets simplification of Retrieval-Augmented Generation (RAG) applications where contextual responses from LLMs are driven by data in a vector store.A demo use case is provided involving hotel data; questions about hotels submitted by users are processed by generating embeddings, querying the Elasticsearch vector store, and forming context-rich prompts for Azure OpenAI.The connector is optimized for .NET developers, allowing them to leverage high-level abstractions for vector storage, search, and prompt design with little code overhead.A detailed technical workflow includes adding NuGet packages, defining a domain model schema, initializing Semantic Kernel, registering AI services and vector storage, and ingesting demo data.The Semantic Kernel engine orchestrates embeddings generation, Elasticsearch search, and prompt injection, all in a modular setup which could easily enable swap-in of alternative AI or data storage services for production scenarios.The RAG pattern illustrated ensures relevant grounding for generative answers, with data and technology choices clearly articulated.The solution addresses pain points in scaling AI-powered enterprise workflows, promoting flexibility and reduced complexity for real-world deployments.
3.5Innovativeness3.5/5Advanced3.5/5 - Advanced. This is a practical enterprise RAG workflow with temporal metadata and table handling, but it is still a domain-specific document QA pattern rather than a highly novel architecture. Compared with nearby calibrated cases like EPA’s document processing journey and Verisk’s RAG assistant, it sits at a similar applied-innovation level.
Deltek collaborated with the AWS Generative AI Innovation Center on a RAG-based solution for question answering across single and multiple government solicitation documents.The solution processes PDF documents with Amazon Textract to extract text and tables, converts tables to CSV, chunks document sections, generates embeddings with Amazon Titan Embeddings G1 – Text v1.2 on Amazon Bedrock, and indexes content plus metadata in Amazon OpenSearch Service.At query time, the system retrieves relevant chunks with semantic search, enriches prompts with metadata such as release date, and uses Anthropic Claude v2 on Amazon Bedrock to generate answers.
How many rag infrastructure use cases are documented?
The AI Use Case Hub documents 6 real rag infrastructure deployments across 5 industries, with 6 detailed company examples you can browse.
Which industries adopt rag infrastructure the most?
RAG infrastructure is most common in Tech & Comms (33%), Finance (17%) and Consumer & Food (17%).
Which countries lead in rag infrastructure?
United States leads documented rag infrastructure deployments, followed by Global and Switzerland.
What technologies are used for rag infrastructure?
Teams most often build rag infrastructure with Amazon Bedrock, Azure OpenAI and Amazon Bedrock Knowledge Bases.
What AI capabilities power rag infrastructure?
Across the documented deployments, the most common capability patterns are RAG (100%), Agent (33%) and Multi-agent (17%).
What results do companies report from rag infrastructure?
Across the 6 deployments reporting outcomes, companies most often cite new product / capability (100%), speed & agility (83%) and better decisions & insight (33%).
