Published on: 13 April 2026

How audaxes combines internal SAP documentation with large language models (LLMs) – and what this could mean for our customers

Artificial intelligence (AI) is transforming the way organizations work with knowledge. This trend is becoming increasingly concrete in the SAP ecosystem – not only through new products such as the Joule copilot, but also through capabilities already available today via the SAP Business Technology Platform (BTP). At audaxes, we have started to actively explore these opportunities through an initial internal prototype that demonstrates what is technically possible.

The Idea: Making SAP Documentation Intelligently Usable

In day-to-day project work, the same questions often arise: Where exactly is it documented how a specific process is configured in SAP Transportation Management (SAP TM)? Or which customizing setting applies to a specific scenario in SAP Extended Warehouse Management (SAP EWM)?

Traditional search approaches – whether in the SAP Help Portal, internal wikis, or general search engines – often return many results but rarely the precise answer. This is exactly where the concept of Retrieval-Augmented Generation (RAG) comes in: instead of querying a language model (LLM) using only general knowledge, it is combined with a domain-specific knowledge base. As a result, the model does not respond based on general probability, but grounded in actual, relevant documentation.

audaxes Prototype: An SAP TM/EWM Knowledge Assistant

audaxes has developed an initial Knowledge Assistant that demonstrates exactly this principle – specialized for SAP TM and EWM. The architecture consists of four layers:

UI Layer
The user submits a question via a simple HTML interface in the browser – intuitive, without the need to access SAP transactions.

Application Layer
A CAP service (Node.js), developed in SAP Business Application Studio (BAS), orchestrates the entire process: it receives the question, coordinates retrieval, builds the prompt for the language model, and returns the response.

Retrieval Layer
This is the core logic: the user’s question is converted into a semantic embedding. This embedding is compared against a vector database containing our curated SAP documentation. Using cosine similarity, the most relevant text chunks are identified and passed to the LLM.

LLM Layer
The actual response is generated by a large language model (LLM) – in our prototype currently via the Gemini API – based on the selected contextual data and enriched with general language understanding.

The SAP-Native Future: HANA Cloud, AI Core, and Beyond

Our prototype illustrates what is already possible today. The next logical step is migration to fully SAP-native BTP technologies:

Current (local/external)	SAP BTP Native
Local data storage	SAP HANA Cloud
Local similarity search	HANA Vector Search
Gemini API	SAP AI Core
HTML test UI	SAP Fiori / Build Apps

With SAP HANA Cloud and its integrated vector search capabilities, external vector databases become unnecessary – the similarity search runs directly within the SAP landscape. SAP AI Core handles lifecycle management of AI models, including scaling, monitoring, and governance. SAP Build Apps or Fiori provide professional, enterprise-ready user experience.

What This Means for SAP Logistics Customers

The practical use cases include:

• Faster project execution: Consultants and key users can ask complex customizing questions directly and receive answers based on their own project documentation.
• Knowledge transfer within teams: Project documentation, blueprint documents, and internal manuals become searchable and conversational for all team members.
• Onboarding new employees: A domain-trained assistant can support new colleagues in learning SAP EWM or TM processes.
• Hypercare support: In operations, frequent support questions can be pre-qualified or answered directly by an intelligent assistant.

Conclusion: Purposeful Experimentation at audaxes

At audaxes, our core expertise lies in SAP supply chain logistics. That is precisely why we know where AI creates real value in SAP logistics solutions. Our RAG prototype demonstrates that SAP BTP already provides the building blocks required to bring this approach into productive use.

audaxes will continue along this path – driven by concrete use cases from our projects and a clear focus on delivering real value for our customers in practice.

If you would like to learn more or are interested in a joint proof of concept (PoC), please feel free to contact us.

Credit: Image by Freepik