SAP unveils tools to help enterprises build their own gen AI apps

SAP has unveiled new tools to build AI into business applications across its software platform, including new development tools, database functionality, AI services, and enhancements to its Business Technology Platform, BTP.

The news came at SAP TechEd, its annual conference for developers and enterprise architects, this year held in Bangalore, the unofficial capital of India’s software development industry.

At last year’s TechEd, SAP unveiled Build, unifying all its low-code applications in a suite of tools for citizen developers. This year, professional developers are the ones being treated to a unified suite of new and existing tools — SAP Build Code.

There’s a common theme to many of SAP’s announcements: enabling enterprise access to business-friendly generative AI technologies.

“We want to accelerate and get generative AI technologies to customers,” said Walter Sun, SAP’s new global head of AI, in a conference call ahead of TechEd.

SAP doesn’t want to build those tools from scratch itself: “We definitely want to leverage what’s already out there,” Sun said, noting there are already many large language models (LLMs) it can build on, adding its own prompting, fine tuning, and data embedding to get those models to business customers quickly.

At the same time, he said, SAP is investing in building a “large business model” — like an LLM, but trained on business transactions — that the company is uniquely placed to build. SAP is training its model on anonymized data from thousands of enterprises using its software, with their consent.

“Other companies don’t have the business data at their fingertips to build such a model,” Sun said. “These large models require a lot of training data. That’s a big investment effort that we’re making at SAP.”

Those initiatives will be made available to users of the new SAP Build Code, among other tools.

SAP Build Code

The new suite will bring together a number of SAP’s existing design and run-time services when it becomes generally available in early 2024. It will be optimized for development in Java and JavaScript, although it’ll also interoperate with SAP’s proprietary ABAP cloud development model, and will use SAP’s Joule AI assistant as a coding copilot.

SAP Build Code will share more than just part of its name with SAP Build, the low-code development suite for citizen developers the company introduced at last year’s TechEd. They both use the same unified application lifecycle management software and governance capabilities, and there’s deep interoperability between the two, the company said, streamlining collaboration between business and IT teams.

The market may not be calling for this interoperability yet, but it’s just as well to plan ahead, said Michele Rosen, IDC’s research manager for low-code, no-code, and gen AI developer technologies.

“We don’t have a lot of evidence indicating that full-time and line-of-business developers are working together closely and simultaneously to build digital solutions,” said Rosen. However, she added, “SAP is right to say that the tools used by different developer personas need to work together.”

The real benefit may be in the governance capabilities rather than the collaboration.

“For low-code development to have real impact on an organization, low-code developers need managed access to core systems controlled by IT, and full-time developers need to be able to work in and with the low-code platform to customize existing components and create new ones,” Rosen said.

SAP is pushing its Business Technology Platform (BTP) as the place for enterprises to build their applications, whether using gen AI or not, rather than hooking them directly into its S/4HANA ERP system. It’s published two new resources for using BTP — a guidance framework with methodologies and reference architectures, and a developers’ guide including building blocks and step-by-step guides — and released an open-source SDK for building extensions on BTP. It’s also adding new logging and telemetry tools to help developers fine-tune their applications.

Plus, there’s a new reward program to encourage enterprises using S/4HANA Cloud private edition to avoid customizing the core of their ERP system, adding extensions on BTP instead. Until now maintaining a “clean core” was considered its own reward, with benefits including easier annual upgrades and simplified system maintenance, but now SAP is offering to reward enterprises with additional credits for BTP usage.

Much of SAP’s focus now is on helping enterprises incorporate gen AI into their enterprise applications, drawing on the ability of LLMs to summarize, synthesize, or generate content.

Vector database

LLMs alone can only answer questions about the data they were trained on — and because the training process is relatively slow and costly, this can’t be kept up to date. They can, however, be enhanced by prompting them with relevant real-time data.

The challenge is obtaining that relevant information from an enterprise’s database, as classical database engines respond with records that are an exact match for a keyword, while relevant information may only match a broad concept — documents might refer to a product enhancement or improvement rather than an upgrade, for example. The solution for gen AI application designers is to use a database optimized for vector search, where a vector is a series of numbers describing the data along different dimensions.

SAP’s other big announcement is it’s enhancing HANA Cloud, the database engine that underpins its S/4HANA ERP system and many of its other applications, to support vector storage and search for unstructured data.

“It takes any kind of business data and embeds that onto a multidimensional vector,” said Jürgen Müller, SAP executive board member for technology and innovation. “This helps put things close together if they are similar.”

This means enterprises building gen AI applications on SAP’s platform can use HANA Cloud to embed relevant up-to-date business information into the prompts they send to LLMs.

“Many companies have their most important data in HANA Cloud already, so they don’t need to deploy a separate vector database for these capabilities,” he said.

Access to HANA Cloud’s vector search won’t be limited to those building on SAP applications; the database is available in all three main hyperscalers’ clouds, and the company said its new gen AI hub enables developers to access a range of LLMs from different providers.

Release of the vector search enhancements, like many of the other new products and features unveiled at TechEd 2023, is planned for Q1 2024.