Case study: RAG for the IT team of Bell Food Group

Case study of a retrieval-augmented-generation (RAG) architecture. It embeds a LLMs and a fine-tuned word embedding model (bge-small).

LLMs en Entreprise

Retrieved Augmented Generation (RAG) for IT Knowledge access


1 - Main Challenge


Documentation and recorded information is crucial for a company, as employees come and go and is therefore needed, to share knowledge between employees.

But often documentation as articles, protocols and records are spread across three different platforms: an internal Knowledge base, an IT knowledge base and a public one.

Employees should have an easy task searching for information, even if the content is not directly posted on the platform itself, but rather given as reference.


2 - How we helped


But can you teach an AI model to help employees find information easily ?

Besides all the internal documentations, we had to create a unique way of creating easy access to information but just asking a question on a document.

We created an AI model that automatically matches the question with the right source(s) of information across the 3 different platforms and summarizes it into an insightful answer for bell food employees.


3 - Technical solution


We used a retrieval-augmented-generation (RAG) architecture. It embeds a LLMs and a fine-tuned word embedding model (bge-small) in order to retrieve the information chunk with a higher accuracy.

4 - Impact


Anabelle Klusmann, Head of Enterprise Architecture and Applications at Bell Food Group AG
Anabelle Klusmann

Head of Enterprise Architecture and Applications at Bell Food Group AG


“The sigmapulse team demonstrated their AI expertise at the 2023 Basel Hack by developing RAG for our IT team at Bell Food Group. With this AI-based solution, it was really easy to access the information of the Bell IT Knowledge Base. Although crafted for the hackathon, its impact on information retrieval accuracy was noteworthy.”


5 - Demonstration



The software was rigorously tested by Anabelle Klusmann and Christine Kerwin, ensuring its effectiveness and reliability.

Both Anabelle and Christine, experts at Bell Food Group AG, played key roles in validating the software's performance and usability.

Anabelle Klusmann, Head of Enterprise Architecture and Applications at Bell Food Group AG
Anabelle Klusmann

Head of Enterprise Architecture and Applications at Bell Food Group AG

Christine Kerwin, Head of Application Management at Bell Food Group AG
Christine Kerwin

Head of Application Management at Bell Food Group AG


7 - Pulse Partners


This project was developed in the context of the Basel Hack competition.


Initial Phase: We recommend the use of GPT-3.5 or GPT-4 for the initial Proof of Concept (POC) phases due to the simplicity of their implementation.


Production Phase: After the POC phase, we suggest adopting LLama2 or Falcon7B, using platforms such as Azure for simplified integration and management. This approach ensures a smooth transition from the testing phase to large-scale production deployment.


Customization: We can also deploy and manage these models directly in your environment, offering a tailored solution to meet your specific needs.


For assistance with your first MVP or a production-ready data system, you can reach out to Pulse Partners: Pulse Partners website.

Stay in the loop

Join our newsletter to get top news before anyone else.