GenAI, but safe
26.03.2025
Together with Ergon, LLB has conducted a proof of concept for the secure use of generative artificial intelligence (GenAI) with sensitive data on Microsoft Azure. The aim was to evaluate the technical capabilities and performance of Large Language Models (LLM) and Retrieval Augmented Generation (RAG) on proprietary data. To ensure compliance with the high security requirements of LLB, Ergon used its GenAI security blueprint, which enabled the services in Microsoft Azure to be comprehensively protected and securely connected. This ensured data control, data protection and data security at all times. The result can be used for a range of use cases.
The LLB is the bank with the longest tradition in Liechtenstein and offers its clients comprehensive wealth management services. As a secure investment partner, the LLB has grown continuously and now has over 1,000 employees in several countries.
The LLB's growth poses particular challenges for the HR department. The number of enquiries is constantly increasing. A lot of time is spent on short knowledge questions - which means less time for solving more complex problems.
As part of a proof of concept, LLB and Ergon wanted to investigate the capabilities of large language models (LLMs) in relation to answering such HR-specific questions. With Retrieval Augmented Generation (RAG), internal data pools can be searched and prepared for the user with the help of an LLM. The resulting demo chatbot was able to answer short knowledge questions from employees.
However, the system requires access to internal data for this. For LLB, it was clear that such use cases had to take place on a secure and protected platform where data control, data protection and data security were guaranteed at all times. To this end, Ergon relied on its GenAI security blueprint.
A secure platform for testing GenAI solutions
With the rapid spread of GenAI tools such as large language models, the question arises as to how they can be used safely and responsibly in order to offer employees and customers added value. Not only must data security be guaranteed, the model should also provide correct answers and be based on facts.
The Retrieval Augmented Generation (RAG) approach offers the possibility of generating answers based on existing company data. This increases the accuracy of the model and it is possible to provide the model with information that it has not previously seen in its training material. It is important that each statement is also provided with a source reference for the purpose of traceability.
As RAG is based on existing company data, an important aspect of the Ergon GenAI security blueprint is the removal of personally identifiable information (PII Removal). This ensures that the privacy of users is protected and that no sensitive data appears in the generated content. Systematic testing with “gold standard” answers is also essential to ensure the quality and accuracy of the generated content. These tests make it possible to evaluate and continuously improve the performance of the models. Safety prompts and content filters help to avoid inappropriate or harmful content and ensure that the content generated is safe and responsible. The technical implementation of the cloud infrastructure with the help of VNETs (Virtual Network) helps to isolate the individual services - such as search and LLM - from each other. By isolating network connections and controlling data traffic, data integrity is protected and unauthorised access is prevented.
With the Ergon GenAI security blueprint, LLB was able to test new models and use cases in a protected environment in the shortest possible time without compromising on security.
Safe, widely applicable
The proof of concept was able to demonstrate that GenAI can be used successfully with sensitive data on a public cloud. The approach can be easily adapted to new business cases.
In just six weeks, Ergon impressed with its diverse expertise: from the development of secure infrastructure in the Azure Cloud, to document intelligence and ETL processes, to prompt engineering with the latest tools and language models. Everything from a single source by our full-stack data scientists.
Do you also want to create business benefits with GenAI and gain experience with the latest methods? Ergon advises and supports the entire process, from designing a data strategy and setting up a data foundation to the successful implementation and operation of your (Gen)AI case.