Building a Simple Chat App with SAP Build Apps and AI Core Document Grounding

Great news, SAP has released the grounding management in AI Launchpad. The grounding management app allows you to manage your data pipelines. You’re able to create a new data repository, connect the document store (e.g., SharePoint) and use the grounding management with the orchestration service.

As someone who enjoys learning by building, I saw this as the perfect opportunity to dive in. I wanted to understand how to setup and use AI Core document grounding, and build a simple chat app using SAP Build Apps to visualize the result. Just a fun, hands-on way to explore the tech and see what’s possible.

This is useful if you want to ensure your model responds based on trusted, internal content rather than generic information.

Connect to data repository

Documents are transformed into vector representations and stored in a vector engine.  The Pipeline API handles everything else for you automatically from preprocessing, chunking and vectorisation.

Alternatively, you could also provide pass chunks directly and store data in the vector database with Vector API.

Here’s how I did it: 

1. Create SharePoint Client Secret

To connect your SharePoint data to the document grounding service, you need to create a secret. This securely stores the credentials needed to access your data source using the OAuth2ClientCredentials.

Assuming you have admin access to Entra and SharePoint, log in to Microsoft Entra, navigate to App registrations, and select your application.

Go to Certificates & secrets in the sidebar, then under Client secrets, click + New client secret to generate one. Take note, the value will be your clientSecret (remember to store it carefully, it will be gone once the page is refreshed).

2. Connect SharePoint to GenAI Hub

Create a secret in SAP AI Core. This secret will be used by the grounding pipeline to authenticate and access the documents.A guide for setting this up here: Create Generic Secret for AI Data ManagementYour credentials (like client_id, client_secret, tenant_id, etc.) need to be Base64 encoded before being added to the secret.Notes: clientId in OAuth2ClientCredentials refers to Application (ClientID) in Azure clientSecret is the secret created in step 1

Once the secret is created, the grounding service can use it to connect to your SharePoint site, pull the documents, and automatically preprocess and index them into the vector engine.

3. Create pipeline in GenAI Hub

From the UI, go to ‘Grounding Management’ tab, click on ‘Create’ to create a new data repository. Details can be found here.From API, send a POST request to the endpoint $AI_API_URL/v2/lm/document-grounding/pipelinesYou can link each pipeline to different folders in your SharePoint for grounding

For a detailed walkthrough, just follow this awesome tutorial to set up your document grounding: https://developers.sap.com/tutorials/ai-core-orchestration-grounding.html

Simplified with Orchestration Workflows

Setting this up was made much easier thanks to the Orchestration Workflow in AI Core. With this, I was able to combine various modules together (more details about each module can be found in the help docs here) :

Integrate data sourcesSelect content filters and craft the promptsChoose the LLM model for retrieval and response generation

All of this can be tested directly through AI Launchpad under Orchestration, which provides a nice interface to see how your model behaves with real queries. There’s a response and trace tab to check the JSON response.

Want to See Where Answers Come From?

AI Core provides a Retrieval API to let you see chunks of content that the model is referencing in its response – a great feature for explaining and debugging. This is great if you want to build your own applications and reference the source document.

Visualizing with SAP Build Apps

To bring everything together, I created a simple chat interface using SAP Build Apps. This frontend lets users type in questions. Everything was built using low-code components, which made it fast to prototype and connect with the backend service.

Deploy your orchestration service to get the URL for consumption.

This tutorial is a great starting point to create your own chatbot with Build Apps. I’ve adapted it slightly to fit the context of document grounding.

If you’re exploring how to bring AI into your applications, this is a great way to get started with SAP. Using SAP Build Apps together with the document grounding service opens up possibilities for building more intelligent applications —powered by your own data.

Thank you to everyone who supported and contributed to this — your insights and guidance were invaluable.

I’d love to hear what you’re working on or thinking about. Feel free to drop any questions or thoughts in the comments!

​ Great news, SAP has released the grounding management in AI Launchpad. The grounding management app allows you to manage your data pipelines. You’re able to create a new data repository, connect the document store (e.g., SharePoint) and use the grounding management with the orchestration service.As someone who enjoys learning by building, I saw this as the perfect opportunity to dive in. I wanted to understand how to setup and use AI Core document grounding, and build a simple chat app using SAP Build Apps to visualize the result. Just a fun, hands-on way to explore the tech and see what’s possible.This is useful if you want to ensure your model responds based on trusted, internal content rather than generic information. Connect to data repositoryDocuments are transformed into vector representations and stored in a vector engine.  The Pipeline API handles everything else for you automatically from preprocessing, chunking and vectorisation.Alternatively, you could also provide pass chunks directly and store data in the vector database with Vector API.Here’s how I did it: 1. Create SharePoint Client SecretTo connect your SharePoint data to the document grounding service, you need to create a secret. This securely stores the credentials needed to access your data source using the OAuth2ClientCredentials.Assuming you have admin access to Entra and SharePoint, log in to Microsoft Entra, navigate to App registrations, and select your application.Go to Certificates & secrets in the sidebar, then under Client secrets, click + New client secret to generate one. Take note, the value will be your clientSecret (remember to store it carefully, it will be gone once the page is refreshed).2. Connect SharePoint to GenAI HubCreate a secret in SAP AI Core. This secret will be used by the grounding pipeline to authenticate and access the documents.A guide for setting this up here: Create Generic Secret for AI Data ManagementYour credentials (like client_id, client_secret, tenant_id, etc.) need to be Base64 encoded before being added to the secret.Notes: clientId in OAuth2ClientCredentials refers to Application (ClientID) in Azure clientSecret is the secret created in step 1Once the secret is created, the grounding service can use it to connect to your SharePoint site, pull the documents, and automatically preprocess and index them into the vector engine.3. Create pipeline in GenAI HubFrom the UI, go to ‘Grounding Management’ tab, click on ‘Create’ to create a new data repository. Details can be found here.From API, send a POST request to the endpoint $AI_API_URL/v2/lm/document-grounding/pipelinesYou can link each pipeline to different folders in your SharePoint for groundingFor a detailed walkthrough, just follow this awesome tutorial to set up your document grounding: https://developers.sap.com/tutorials/ai-core-orchestration-grounding.html Simplified with Orchestration WorkflowsSetting this up was made much easier thanks to the Orchestration Workflow in AI Core. With this, I was able to combine various modules together (more details about each module can be found in the help docs here) :Integrate data sourcesSelect content filters and craft the promptsChoose the LLM model for retrieval and response generationAll of this can be tested directly through AI Launchpad under Orchestration, which provides a nice interface to see how your model behaves with real queries. There’s a response and trace tab to check the JSON response. Want to See Where Answers Come From?AI Core provides a Retrieval API to let you see chunks of content that the model is referencing in its response – a great feature for explaining and debugging. This is great if you want to build your own applications and reference the source document. Visualizing with SAP Build AppsTo bring everything together, I created a simple chat interface using SAP Build Apps. This frontend lets users type in questions. Everything was built using low-code components, which made it fast to prototype and connect with the backend service.Deploy your orchestration service to get the URL for consumption.This tutorial is a great starting point to create your own chatbot with Build Apps. I’ve adapted it slightly to fit the context of document grounding. If you’re exploring how to bring AI into your applications, this is a great way to get started with SAP. Using SAP Build Apps together with the document grounding service opens up possibilities for building more intelligent applications —powered by your own data.Thank you to everyone who supported and contributed to this — your insights and guidance were invaluable.I’d love to hear what you’re working on or thinking about. Feel free to drop any questions or thoughts in the comments!    Read More Technology Blogs by SAP articles 

#SAP

#SAPTechnologyblog