Advancing knowledge retrieval with Large Language Models

Linde, a leading global industrial gases and engineering company, required assistance in developing an internal tool to facilitate knowledge retrieval across their multi-national organization. The goal was to design a simple, top-quality, and secure solution that could quickly and reliably answer everyday work-related questions.

Addressing the challenges of knowledge retrieval

Linde sought a partner experienced in Machine Learning technologies who could implement a custom Large Language Model (LLM) to enable chatting with the company’s assets, such as PDF files, while guaranteeing a secure and reliable process.

To meet this need, STX Next integrated a Retrieval Augmented Generation (RAG) tool focused on quick, easy, and expert knowledge retrieval. By utilizing a database of the company’s documents, the solution made information readily accessible to all users. This approach ensured smooth operation and significantly reduced the time required to find information.

Integrating the
Large Language Models-powered tool

Integrating the Large Language Models-powered tool

After a thorough analysis of the customer's needs, our dedicated team proposed the implementation of new AI-based tools and features, including:

Retrieval Augmented Generation tool

Combining document retrieval with LLM generation allows the language model to provide relevant information. First, the model searches the document database for the answer and then responds directly to the user.

Proprietary data security

The solution was hosted on Azure Cloud, keeping data secure throughout the process and not transferred to external organizations.

Multi-national support

The solution was designed to be multilingual from the start, allowing user interactions from various regions worldwide.

Optimized document indexing

The indexing and retrieval process was optimized by testing text-splitting methods, adding search-friendly metadata, and handling tables separately. We also developed automated processes to keep the database updated as data sources evolved.

Maintaining reliability

The system model cites its sources, enabling users to verify information quickly. Additionally, we implemented a framework to test and monitor the chatbot's performance.

Looking to leverage modern technologies for growth?

Share your vision with us and discover the perfect solution for your needs.

Get started

Faster, adequate information searches

The incorporation of these features resulted in significant benefits for Linde and their employees in everyday work:

Improved information search

Users can now quickly find answers to questions about internal policies, eliminating the need to sift through multiple search engines or consult colleagues.

Broader information access

Thanks to its multilingual capabilities, the system allows users to ask questions in one language and find answers in documents written in another. This enables access to information that was previously out of reach.

Enhanced comprehension

The natural language interface allows users to ask follow-up questions or request additional summaries, ensuring they fully understand the answers provided.