This blog is part of the series Generative AI with SAP, which is focused on understanding how SAP is leveraging AI and generative AI across its portfolio.
In this series we are learning:
To make it easy to understand, the blog series is divided in small parts. Each blog requires maximum 10–15 minutes to learn.
This is the 4th blog in the series, where will learn about AI Foundation.
Note: I will publish the subsequent blogs soon.
In January 2021, SAP introduced a unique and powerful solution offering called SAP Business Technology Platform (SAP BTP). SAP BTP is the technical foundation of entire SAP ecosystem and plays a crucial role for all SAP customers and partners.
In other word, SAP BTP is a portfolio of SAP solutions and services that are brought under one umbrella. These services, and solutions helps organizations build a new cloud solution or extend SAP systems.
For more information on SAP BTP, see Explaining SAP Business Technology Platform (SAP BTP) to a Beginner
Imagine you are developers/architect working for an SAP customer or partner.
AI foundation is an offering from SAP, which fulfil all these requirements.
AI foundation on SAP BTP is a set of ready-to-use AI services and tools, which:
In other words, AI Foundation is the set of AI services and tools on SAP BTP for developers to incorporate AI capabilities into applications/extensions.
Below image gives a high-level picture of AI foundation.
SAP has partnered with leading general-purpose AI vendors and large language model (LLM) providers to ensure that SAP customers are able to reap the benefit of AI innovations without going into individual contract with 3rd party vendors.
For example, if you need to use GPT LLM, you do not need to buy any separate license, you can access it via Generative AI Hub on SAP BTP. From Generative AI Hub, you can access these large language models from multiple vendors easily.
Below image, shows the major components of AI Foundation.
Let’s look into each of these components in detail.
AI services (aka AI Business Services) are ready-to-use, reusable AI services on SAP BTP.
These AI services are broadly categorized in 3 buckets as shown in below image.
AI foundation provides a ready-to-use service Document Information Extraction for intelligent document processing.
The Document Information Extraction service:
With Document Information Extraction service, we can:
We can use the extracted information to automate business processes. For example, automatically process payables, invoices, or payment notes while making sure that invoices and payables match.
There are 3 services under this bucket to enhance human-decision making:
Personalized Recommendation is a generic reusable service, which:
For example, SAP is leveraging this service to process more than 4.2 million learning recommendations for SAP SuccessFactors customers each month.
Data Attribute Recommendation service can be used if there are missing values in the dataset. The service:
With Data Attribute Recommendation we can:
Business Entity Recognition is a service which helps to detect and highlight any type of business entity in unstructured text. For example, this service can be used to automatically extract the context from incoming emails with invoice inquiries, automating recurring tasks associated with answering queries about the status and payment of invoices.
SAP is being used all over the world and in almost all the languages and that brings an important aspect of Translation of Texts to all the required languages. To accelerate and automate the translation process, SAP offers SAP Translation Hub service under AI Foundation.
SAP Translation Hub is a reusable service to translate content into multiple languages. This service helps you speed up the translation of software texts (e.g. user interfaces) and related documents – with high quality and accuracy. You can use a repository of SAP-approved translations and terminology as well as machine translation, which has been trained with focus on SAP-specific content.
Note: You can find the list of all the AI services and their details in SAP Discovery Center or you can visit the official product documentation.
Customers and partner can integrate the SAP AI Services into their own processes and applications by connecting and calling the provided API endpoints accordingly.
Let’s understand this with an example.
Let’s say we want to use Personalized Recommendation service to give visitors to your website highly personalized recommendations based on their browsing history and/or item description.
Below diagram shows the high-level architecture diagram for this use-case.
We can achieve this as a sequence of steps shown below.
Note: You can follow this tutorial to get step-by-step guide for this entire process.
Similar to the approach we took for Personalized Recommendation, we can use other AI services as shown in below diagram.
Use Machine Learning to Extract Information from Business Documents and Enrich Data
Use Pre-Trained Machine Learning Models to Process Unstructured Text
Shape Machine Learning to Process Custom Business Documents
Use Machine Learning to Classify Data Records
Use Generative AI to Process Business Documents
Second major component of AI Foundation is Generative AI Management. Under this bucket, AI foundation provides an offering called Generative AI Hub – which is an BTP service to explore different LLM models.
Let’s have a close look into Generative AI Hub.
Generative AI Hub helps us to fast-track development of any generative AI apps on SAP BTP. It enables developers to instantly access broad range of large language models (LLMs) from different providers, such as GPT-4 by Azure OpenAI or Open-Source Falcon 40B.
Generative AI – an BTP service to explore different LLM models.
We can use Generative AI Hub to orchestrate multiple generative AI models:
The generative AI hub also provides tooling for prompt engineering, experimentation, and other capabilities to speed-up the development of generative AI applications on BTP, in a secure and trusted way.
Below image shows a quick glimpse of how we can submit a prompt to multiple LLMs and compare the generated outcomes to identify the best-suited model using Generative AI Hub.
Disclaimer: The above image is taken from this blog
Note: For more information on Generative AI Hub, refer to Part 9 – Overview of Generative AI Hub [to be published]
AI workload management are targeted for developers who wants to build a new AI model from scratch or run an AI workflow or execute an AI asset in a standardized, scalable, and hyperscaler-agnostic way. AI workload management offers all the necessary tools to create AI models, train them, evaluate their accuracy and publish them for inferencing.
Under AI Workload Management bucket, AI foundation provides two services:
SAP AI Core is a service available on SAP BTP, which offers a powerful AI Runtime. We can use SAP AI Core to train and deploy our AI models cost-efficiently at scale while preserving privacy and compliance.
One of the main benefits of SAP AI Core is that we can choose underlying infrastructure (GPUs, CPU Cores, Memory) from a broad range of plans, as well as avail built-in autoscaling. You can have a quick look on the various resource plan available for SAP AI Core here.
Note: For more information on SAP AI Core, refer to Part 5 – Introduction to SAP AI Core
SAP AI Launchpad is a service on SAP BTP, which can be used to manage and run AI workflows and AI models across multiple AI runtimes (including SAP AI Core).
SAP AI Launchpad provides generative AI capabilities via the Generative AI Hub.
SAP AI launchpad also provides centralize AI lifecycle management for our AI scenarios with a convenient user interface. We can monitor model performance statistics continuously and retrain as needed.
Note: For more information on SAP AI Launchpad, refer to Part 6 – Introduction to SAP AI Launchpad [to be published]
The most important differentiator of SAP system from other AI providers is that – SAP systems contains huge business data of customers. These business data is the key to leverage AI models to solve real business problems.
With solutions like SAP HANA Cloud and SAP Datasphere, SAP helps providing business data and context to the AI models. SAP HANA Cloud vector engine has specifically been designed for generative AI use-case.
The vector engine is a new addition to SAP HANA Cloud, a database-as-a-service delivered on SAP BTP. Using SAP HANA Cloud vector engine, we can ground AI with unique business data.
The SAP HANA Cloud vector engine supports efficient storage and processing of vector data. This means we can execute joins, apply filters, and perform selects by combining vector data with various data types, including transactional, spatial, graph, and JSON data, all within the same SQL environment.
Here are some key benefits of the SAP HANA Cloud vector engine:
Note: For more information on SAP HANA Cloud vector engine, refer to Part 10 – SAP HANA Cloud vector engine [to be published]
SAP HANA Cloud also provide AI function libraries - Predictive Analysis Library (PAL) and Automated Predictive Library (APL). These libraries allow us to implement classification, regression or time series forecasting scenarios applied directly to the business data.
SAP also provides a solution, SAP Datasphere for data management. SAP Datasphere includes AI in several ways, for example in data integration, data quality & cleansing (such as correcting missing values or duplicates), data governance, analytics (such as detecting trends, analyze historical data, and make predictions), and natural language processing.
Under AI Foundation umbrella, SAP BTP provides a wide range of foundation models that can be used to implement various generative AI use-cases. To provide a seamless experience to SAP customers, SAP has partnered with leading general-purpose AI vendors and LLMs providers, such as Google, Meta, IBM, Nvidia etc.
These foundation models, help ensure that SAP customers keep up with the fast pace of innovation and have the flexibility of choosing right model for their scenario.
A foundation model is a deep learning model which is trained on huge amount of data, usually with unsupervised learning. The model can be adapted to perform a wide range of tasks such as text generation, image generation, video generation, sentiment analysis, information extract etc.
Foundation models can be considered as general-purpose technologies that can support a diverse range of use cases. Since these models are pre-trained on vast datasets using powerful hardware and techniques, they can be used to save time and computational costs compared to training new models for each specific application.
Below image summarizes important points about foundation model.
To know more about foundation model, you may refer to What is Foundation Model?
Let’s first understand why customers need a foundation model from SAP.
There are many general-purpose foundation models available which are pre-trained on huge data. Under AI Foundation, SAP provides easy access to many of these models such as GPT-4, Llama2, Falcon-40b, Claude2 and many more. 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 LLM as a colleague who is very smart and knowledgeable. This colleague can answer most of your queries. However, he may not know what has happened in the world last year and he also does not have any idea about your company’s internal process and policies.
For example, let’s say you ask a LLM - “What do you think about the offer from our most important supplier last week?”.
An LLM can only work with the initial training data and may nor provide accurate answer.
However, if you provide enough business data and context to the LLM, then you can get the answer you expect.
Supplementing this lack of information is where SAP Foundation Model comes into the picture. SAP is trying to adapt the LLMs with the SAP business data and context to create it’s own models which will be best suited for customers.
SAP is designing its own foundation models by combining general-purpose large language models with its deep understanding of business processes and business data. These models will help customers to get context-specific response.
The SAP foundation model will be very useful in all core business areas such as finance, sales, supply chain, etc. It will be able to address questions we face every day in business that large language models cannot answer properly, such as predicting invoice payment dates and supplier delivery quality or proposing efficiency improvements to a business process.
SAP foundation model will also allow for fine-tuning for different scenarios, will further unlock and strengthen SAP’s leading position in embedded AI. Customers will benefit from high-performance models trained on rich SAP data, achieving better predictive results.
Please note that customers don’t have to wait for these SAP foundation models to be ready before they can leverage context from business data. There is another approach called Retrieval-Augmented Generation (RAG), which also can be used to provide business data context and give customers better results right away.
Note: For more information on RAG, refer to Part 11 - RAG, a real-life use case using SAP AI Core, Generative AI Hub and HANA Vector Engine [to be published]
Part 5 – Introduction to SAP AI Core
Disclaimer– All the views and opinions in the blog are my own and is made in my personal capacity and that SAP shall not be responsible or liable for any of the contents published in this blog.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
User | Count |
---|---|
21 | |
19 | |
11 | |
10 | |
9 | |
9 | |
7 | |
7 | |
6 | |
5 |