[ Industry case study — AI & Automation ]

Multilingual RAG platform puts Linde's knowledge seconds away

STX Next built a secure, Azure-hosted RAG platform that lets Linde employees query internal documentation in natural language and get source-cited answers in seconds.

Original project by STX Next ↗Industry reference — not a NexSum Labs engagement.

Client
Linde
Industry
Industrial gases & engineering
Services
Retrieval-augmented generation, AI/ML engineering, Data engineering, Cloud infrastructure
Year
2024
STX Next and Linde case study banner showing an engineer in a hard hat holding a tablet at an industrial site
Seconds
to a source-cited answer, replacing manual document searches
Cross-language
querying across multilingual documents
Azure-hosted
all data stays inside Linde's own infrastructure

The client

Linde is a leading global industrial gases and engineering company with a large international workforce. It holds a vast library of internal technical, policy and operational documentation written in many languages.

The challenge

Information on equipment specifications, safety protocols, maintenance procedures and internal policies was scattered across PDFs, scanned files, tables and multilingual content. Finding accurate answers was slow, error-prone and meant searching through several systems. Linde wanted a simple, secure way for employees to chat with the company knowledge base in natural language, wherever they are.

The solution

01Custom RAG pipeline

User questions first retrieve the most relevant passages from Linde's document store, and an LLM then generates a contextual answer grounded in those passages.

02Data sovereignty on Azure

The whole solution is hosted on Microsoft Azure inside Linde's own infrastructure, with no external data transfers.

03Multilingual document handling

Advanced text splitting, search-friendly metadata, separate table handling and automated sync pipelines keep a mixed corpus of PDFs, scans and tables indexed and current.

04Citations and quality monitoring

Every answer cites the exact documents and passages it came from, and a monitoring framework continuously tracks chatbot reliability.

The results

Employees now get accurate answers in seconds instead of digging through shared drives or asking colleagues. Cross-language retrieval opened knowledge to staff in every region regardless of the source document's language. Source citations and in-house hosting increased trust, and automated indexing keeps the knowledge base maintainable as documentation grows.

Tools & stack

PythonFastAPIElasticsearchOCRAzure Document IntelligenceMicrosoft AzureLarge language models

Original project by STX Next ↗ — published by the delivering agency, included here as an industry reference.

Next case study — CometChat

CometChat redesign drives 50% growth in demo requests