Power your business with secure and scalable generative AI services from AWS and SAP

Introduction

The ability to adopt and innovate with cutting-edge technologies is not a luxury but a requirement for organizations that want to position themselves for continuous success. Thousands of customers run business-critical SAP workloads on Amazon Web Services (AWS), and they tell us they’re seeking new ways to build innovative solutions to enhance their business operations. With the recent advent of mainstream generative artificial intelligence technology, our SAP customers are looking to leverage Amazon generative AI services such as Amazon Bedrock to accelerate innovation within their business processes such as Order to Cash, Procure to Pay, Hire to Retire, Design to Produce etc., This is why earlier this week, AWS and SAP announced the expansion of their strategic partnership, including new capabilities for generative AI. Amazon Bedrock models are now available in SAP generative AI Hub within SAP AI Core. This will enable enterprises leveraging SAP ERP to build generative AI-powered application extensions using SAP Business Technology Platform (SAP BTP) and foundation models (FMs) from Amazon Bedrock while continuing to maintain a clean core. This blog provides technical guidance from AWS and SAP in the form of a ‘Joint Reference Architecture’ (JRA) to our customers who are looking to consume FMs from Amazon Bedrock within their SAP ecosystem.

Building with generative AI on AWS

While customers seek guidance on how generative AI supports their enterprise-wide modernization strategy, they also expect these solutions to meet the high bar of security and infrastructure reliability for their SAP workloads. To help customers modernize, automate and innovate with generative Gen-AI technology, we launched Amazon Bedrock as a fully managed service which provides a wide range of high-performing foundation models (FMs) from leading AI companies such as Anthropic, AI21 Labs, Cohere, Stability AI, Mistral, Meta Llama, and Amazon Titan. Amazon Bedrock offers a comprehensive set of capabilities to build Gen-generative AI applications, simplifying development while maintaining privacy and security. Key features include model customization with your own data, fine-tuning for specific tasks, and Retrieval Augmented Generation (RAG) to enhance response accuracy using your company’s knowledge base. Bedrock also supports building intelligent agents that can automate tasks by interacting with your enterprise systems. With robust security measures, including data encryption and compliance with standards like SOC and ISO, Amazon Bedrock ensures a secure environment, making it an ideal choice for developing innovative AI-driven applications.

AWS offers other generative GenAI services infrastructure like AWS Trainium and AWS Inferentia to enable customers to train their AI models and run inference on the cloud effectively at lower cost. Additionally, developers can leverage Amazon Q Developer, an AI coding companion to improve developer productivity by generating code suggestions real-time, based on developers’ comments in natural language and prior code in their Integrated Development Environment (IDE).

AWS and SAP Partnership

SAP and AWS have been partnering since 2008 to help our customers run their SAP applications better and innovate faster. SAP and AWS have joined forces to come up with a set of Reference Architectures to tackle practical business scenarios under the modernization umbrella, bringing the power of AWS Services to SAP customers through SAP Business Technology Platform (SAP BTP). By embracing a clean core model, these Joint Reference Architectures (JRA) provide a framework for application development and an integrated extension to SAP S/4HANA to achieve business process automation & optimization. This JRA built by SAP and AWS provides joint-expert guidance to customers for building new scalable applications, analytical dashboards, or machine learning models for our customers. AWS and SAP also plan to adapt the JRA to new services and features released by SAP and AWS in the future.

Integrating generative AI models from Amazon Bedrock with SAP – Joint Reference Architecture Overview

SAP has been an early adopter of generative AI technology, with the release of the Generative AI hub on SAP BTP. The Generative AI hub in SAP AI Core provides purpose-built AI development tools, enterprise-grade access to leading AI foundation models, and robust data control for creating AI-powered applications. The integration of generative AI models from Amazon Bedrock through the Generative AI Hub in SAP AI Core, SAP customers can access family of foundation models such as the Anthropic Claude3 and Amazon Titan. With this joint reference architecture, SAP customers can accelerate the adoption of generative AI and modernize key business processes built on SAP solutions. These innovations can be used in embedded use cases within RISE with SAP and the intelligent scenario lifecycle management functionality as an integration component or side-by-side directly on SAP BTP. Customers can use Generative AI hub in SAP AI Core and AWS services to build generative AI solutions to further enable custom AI functionality within SAP’s portfolio of cloud solutions and applications. This can help deliver new insights and optimization across various business functions including Finance, Human Resources and others. SAP and AWS plan to expand the use of Amazon Bedrock capabilities in the Generative AI hub to further enable embedded AI functionality within SAP’s portfolio of cloud solutions and applications. This includes use cases across finance and product lifecycle management.

Let’s take a closer look at the architecture and the various components involved.

Amazon Bedrock: Amazon Bedrock allows customers to choose from a wide choice of Large Language Models (LLMs), that will be made available through APIs. In this JRA, we use Amazon Titan and Anthropic’s Claude that are a family of Foundational Models (FMs) pre-trained on AWS with large datasets, making them powerful, general-purpose models built to support a variety of use cases. Use them as is or privately customize them with your own data.

Generative AI Hub in SAP AI Core: SAP AI Core service exposes AI assets such as large language models to customers and provides unified interfaces for SAP applications that run on the SAP BTP ecosystem. In this JRA, we use the Generative AI hub in SAP AI Core as an access and lifecycle management layer to manage access to Amazon Bedrock and present an endpoint for our application to consume the foundational models. Through the Generative AI Hub, SAP centrally enforces numerous content filtering, SAP-specific risk mitigation, and safety guardrails, providing a compliant approach to safeguard against potential business and legal risks at scale across the SAP ecosystem.

The Generative AI hub in SAP AI Core gives developers instant access to a broad range of large language models (LLMs) from different providers through a managed commercial and legal framework. With this access, developers can orchestrate multiple models. The Generative AI hub will also connect to the vector capabilities in SAP HANA Cloud to help developers reduce model hallucinations and incorporate contextual data as embeddings to deliver more tailored results to specific use cases.

Data storage and retrieval: SAP HANA Cloud is a multi-model database management system that helps to build and deploy intelligent data applications at scale. SAP HANA Vector engine can support Retrieval Augmented Generation (RAG) to obtain better results from LLMs.

Application development: Cloud Application Programming (CAP) Model is one of the approaches to develop cloud applications by using SAP Build. CAP provides a more structured and seamless framework for data modeling and enhanced integration with other services. CAP offers developers a variety of open source and SAP frameworks to streamline development through accelerated innovation. In this JRA we are using CAP as the entity layer of an application fronted by a SAP UI5 front end.

Authentication and Identification: Large language models require as much data as possible but ignore user authorizations at query time, making integration challenging. To circumvent this,we will use SAP BTP and SAP Identity Provisioning services to manage SAP and non-SAP identity lifecycles for model prompting.

With the consolidated use of these components, customers will now be able to build a full stack generative AI powered SAP application leveraging Amazon Bedrock and SAP BTP services that is scalable and reliable. This JRA pattern can not only be adapted to a wide range of business process extensions within an enterprise-grade SAP landscape, but also be helpful in maintaining SAP recommended clean core approach.

Conclusion

In this blog, we have discussed guidance from SAP and AWS in providing a reference architecture to our customers for consuming generative AI capabilities of Amazon Bedrock with SAP BTP. Using this architecture, SAP workloads can now be supplemented with large language models (LLMs) and transformer model capabilities to harness the power of SAP data, resulting in improved insights and operational efficiencies at a lower cost. SAP and AWS are now working towards defining specific use cases that can make use of Amazon Bedrock’s generative AI capabilities via SAP BTP to solve business problems and improve customer experience through automation and innovation. Stay tuned for more.

Check out the following blogs the joint teams published in 2022-2023 with respect to the joint reference architectures for SAP workloads on AWS.

SAP and AWS – Joint Reference Architectures to maximize utilization and investments
AWS and SAP – Joint reference Architecture for IoT scenarios using Amazon Monitron

You can find out more about AWS for SAP, Amazon Bedrock from the AWS product documentation. If you require additional expert guidance, contact your AWS account team to engage a local SAP specialist solution architect or the AWS Professional Services SAP specialty practice.

Join the SAP on AWS Discussion

In addition to your customer account team and AWS Support channels, we have launched re:Post where our AWS for SAP Solution Architecture team regularly monitor the AWS for SAP topic for discussion and questions that could be answered to assist our customers and partners. If your question is not support-related, consider joining the discussion over at re:Post and adding to the community knowledge base. To learn more about thousands of active customers run SAP on AWS, visit the AWS for SAP page.

Credits

The AWS and SAP partnership on Joint Reference Architectures are the result of deep collaboration and contribution from SAP and AWS organizations. We would like to thank the following members for their expertise, support and guidance.

• Team AWS: Sunny Patwari, Yuva Athur, Ganesh Suryanarayanan, Spencer Martenson, Steve DiMauro and Soulat Khan.
• Team SAP: Madankumar Pichamuthu, Weikun Liu, Daniel Zhou, Sivakumar N and Anirban Majumdar