The new SAP HANA Cloud vector engine enables businesses to combine the power of large language models (LLMs) with company-specific, real-time data and business process know-how, all integrated in one multi-model database: SAP HANA Cloud. With the latest quarterly release, the vector engine is now generally available.

SAP HANA Cloud is a market-leading database-as-a-service enabling intelligent data applications and is one of the most adopted services within SAP Business Technology Platform (SAP BTP) internally at SAP. As of today, more than 180 different applications and services use SAP HANA Cloud with its multi-model capabilities.

Now, SAP HANA Cloud is also a leader in the generative AI age.

At SAP, we work with various LLMs such as GPT-4, Llama2, Falcon-40b, and Claude2. While these models offer amazing opportunities, they also have limitations. For example, LLMs may rely on outdated training data and lack company-specific data and business process context.

As an example, imagine having an LLM as a colleague. This colleague would be very intelligent, able to program, pass exams, or have arguments – but this colleague would not know anything about what happened in the world in the past year, nor have any idea about internal processes of your company or any of your systems. Even worse, after every conversation you have, this colleague would forget what you just talked about. Working with such a lack of memory would be of limited value. This shortcoming is why an LLM cannot answer easy questions like “What do you think about the offer from our most important supplier last week?” An LLM can only work with the initial training data – all other data must be provided as context.

Supplementing this lack of information is where SAP HANA Cloud vector engine can assist. The engine can provide LLMs with all the relevant data of an organization through a process called “retrieval-augmented generation.”

A Game-Changing Feature

So how does the vector engine work? It is a new addition to SAP HANA Cloud’s multi-model engines, enabling customers to utilize the similarity between two or more vectors to solve business problems. With the integration of AI-focused technology, SAP HANA Cloud can now empower businesses to combine intuition along with data-driven insights to solve even the most complex of problems.

Some key benefits and features of the vector engine include:

• Multi-model: Users can unify all types of data into a single database to build innovative applications using an efficient data architecture and in-memory performance. By adding vector storage and processing to the same database already storing relational, graph, spatial, and even JSON data, application developers can create next-generation solutions that interact more naturally with the user.
• Enhanced search and analysis: Businesses can now apply semantic and similarity search to business processes using documents like contracts, design specifications, and even service call notes.
• Personalized recommendations: Users can benefit from an improved overall experience with more accurate and personalized suggestions.
• Optimized large language models: The output of LLMs is augmented with more effective and contextual data.

The Database Foundation of SAP’s Generative AI Strategy

The addition of the vector engine establishes SAP HANA Cloud as the default database in SAP’s generative AI solution strategy. Customers can create the next level of user experiences along with other services within SAP BTP. As an example, SAP BTP can provide centralized access to SaaS-based LLMs from multiple vendors as well as host LLMs from open-source models or third parties. The generative AI hub in SAP AI Core, a capability that facilitates the use of generative AI capabilities, will soon rely on SAP HANA Cloud as the primary vector storage. One function of the generative AI hub feature is to help provide a process for creating embeddings and storing the resulting vectors in SAP HANA Cloud. Customers building intelligent data applications can use both services together to augment LLM queries with relevant context for meaningful answers.

SAP is working on foundation models that are specific for SAP-related industry and process knowledge.

The Database for Innovation

SAP HANA Cloud continues to lead the market by storing and processing different types of relevant business data – all within the same database. The new vector engine, combined with other multi-model capabilities, opens a world of possibilities for applications to help enhance the execution of business processes. Whether improving search capabilities, gaining deeper insights for informed decisions, or optimizing LLMs, SAP HANA Cloud enables the type of applications that can elevate the expertise and effectiveness of every user.

To learn more, sign up for an introductory webinar on April 4. Do you already have a use case for SAP HANA Cloud vector engine in mind? If so, consider registering for the SAP Early Adopter Care program.

Juergen Mueller is CTO and member of the Executive Board of SAP SE, Technology & Innovation.
Stefan Baeuerle is head of Database, SAP HANA Database, & Analytics for Technology & Innovation at SAP.

Carbon footprint calculation, patient privacy, and machine learning based on sensitive data – thanks to advanced encryption methods like fully homomorphic encryption.

Most have been in this situation before: one of the providers or services we use is a victim of a data breach and we want to determine if our personal user data has been impacted. This is where fully homomorphic encryption (FHE) comes into play. With FHE, the encrypted, personal password is compared against the data set of stolen user data and potential matches are identified without ever revealing the user’s password.

Use cases for this type of privacy-enhancing technology (PET) are numerous. They range from applications in medicine, where third-party service providers can analyze health data without compromising a patient’s privacy, to performing machine learning and AI algorithms on encrypted data, allowing organizations to derive insights from sensitive data sets without exposing the data to potential breaches or privacy violations.

How It Works

Fully homomorphic encryption allows calculations to be performed on encrypted data without having to decrypt it first. Confidentiality is maintained, as even the results are encrypted and can be viewed only with the appropriate decryption key. Further techniques for processing encrypted data are multi-party computation (MPC) and trusted execution environments (TEE).

Mathias Kohler, research manager at SAP Security Research, outlines the differences: “While FHE is the most known of the encryption technologies, MPC is the ideal candidate if working with several parties exchanging encrypted data across company borders. And it can be substantially faster than FHE.” While both are software-based technologies, TEE is hardware-based, which makes it the fastest choice. The downside: TEEs, unlike MPC and FHE, require decrypting the data for processing. While decryption happens in a trusted hardware environment isolated from the operating system, it can allow data leakage via side-channel attacks. Notably, PETs do not need to be considered in isolation and can augment each other. For example, MPC can encrypt and distribute an FHE decryption key, protecting the FHE key and ensuring no single party can decrypt everything.

Why It’s Relevant

There is a demand for this kind of technology. By 2025, 60% of large organizations will use at least one privacy-enhancing computation technique in analytics, business intelligence, or cloud computing, according to Gartner.

Fully homomorphic encryption has numerous applications, especially in scenarios where privacy and security are paramount, such as secure computation in the cloud, privacy-preserving data analysis, and secure outsourcing of computations. As long as one party is performing the data processing centrally, FHE is the encryption method of choice. FHE enables organizations to share encrypted data with partners or third parties for analysis or monetization purposes while maintaining data confidentiality. This is particularly relevant in industries such as advertising and market research.

Interesting use case scenarios from SAP’s perspective could be secure benchmarking and predictive maintenance.

Secure Benchmarking

Companies often assess their competitiveness relative to industry peers and compare business-relevant KPIs, such as automation rate or return rates, with peers and even competitors. With fully homomorphic encryption, all participating parties can share encrypted KPIs without revealing individual data. As a result, they learn about relevant statistics, such as averages or medians, to assess their relative competitiveness and decide where to improve and invest.

Predictive Maintenance

Predictive maintenance is a machine learning technique to forecast demand for maintenance or spare parts based on historical data. “In certain industries, required data, such as usage patterns and failures, is considered sensitive and is not easily shared with data scientists or maintenance operators,” says Anselme Tueno, senior researcher at SAP Security Research. By computing on encrypted data, however, no sensitive information is revealed while still allowing for the required insights to be gathered for prediction tasks.

Carbon Footprint Calculation with Multi-Party Computation

While it is early days from a product availability perspective, SAP is working on potential use cases with customers and partners. One key example is calculating carbon footprints of products.

Prime examples for complex collaborations are today’s supply chains, intricate networks that encompass various levels of suppliers, manufacturers, and processed goods. Unfortunately, there is often a lack of comprehensive visibility across the entire process – either for technical reasons or because businesses are often reluctant to share sensitive data across supply chains that often include direct competitors.

However, to accurately assess and disclose a product’s carbon footprint, sensitive production details and associated carbon costs for production-relevant parts and materials are required. Here, MPC can reveal only the required carbon footprint without disclosing associated, proprietary manufacturing details with other supply chain participants.

Currently, SAP is working with Bosch on cloud-native software for secure multi-party computation called Carbyne Stack.

“SAP participates in this open-source project and supports the development of Carbyne Stack’s storage and processing services and the deployment of Carbyne Stack on Amazon Web Services (AWS),” Kohler explains. “For Bosch, Carbyne Stack is a type of cloud-native operating system for MPC workloads that manages resources to run as efficiently as possible in multi-cloud deployments.” This effort can help SAP in the long run to integrate MPC as technology into SAP solutions and services while running in a cloud-native environment.

What’s Next?

Despite all the benefits around processing data, encryption introduces significant computational overhead due to the complexity of performing operations on encrypted data. Slow processing speeds, especially for complex operations and large data sets, makes fully homomorphic encryption impractical for real-time applications or large-scale data processing. Although the performance of FHE has greatly improved in recent years, its practical adoption is still limited due to the processing overhead and performance considerations. Ongoing research is focused on the design of FHE-specific hardware accelerators.

“PETs for computing on encrypted data have the power to amplify data-driven business collaborations and reshape the future of cloud computing,” explains Jonas Böhler, senior researcher at SAP Security Research. By safeguarding data, they enable access to previously untapped information while minimizing privacy risks and thwarting data breaches. The future of computing is encrypted.