Artificial intelligence, which is based on the Large Language Models (LLMs), is now an essential lever for improving productivity. Regardless of the job in which we work, we will soon all interact with virtual assistants. They are there to help us find the information we need more quickly, to verify documents or computer code, to produce summaries or presentations. "Implementing these solutions at the heart of companies is one of the main challenges of the moment," comments Steve Heggen, Head of HyperAutomation at Fujitsu Luxembourg. The integration of artificial intelligence solutions in a professional context, and more particularly within regulated sectors such as finance and insurance, raises many questions, particularly in terms of preserving data confidentiality. »

A solution that can be deployed locally

In this context, in order to allow Luxembourg players to take advantage of the possibilities offered by AI, Fujitsu Luxembourg has chosen to use Devana. This solution facilitates the implementation of virtual assistants adapted to the needs of companies. "One of the great things about this solution is the ability to install it at the customer's IT infrastructure, where users can connect to language models available in the cloud, as well as to high-performance models that can be hosted locally," says Heggen. It is therefore not necessary to expose your data to the outside. Their confidentiality is ensured. »

Indicate the reliability of the answers

Beyond the guarantees provided in terms of data security, the Devana solution responds to other concerns of companies and regulators alike. "In particular, Devana has been endowed with the ability to cite sources related to the answers that a virtual assistant will give to a question. In the interface, the elements advanced by the AI are also accompanied by a score, reflecting the level of reliability of the result delivered," comments Marvin Sant, CTO of Devana.

This feature, more than indicating to the user the level of relevance of the virtual assistant's answers, makes it possible to explain how the latter arrived at the advanced results. "It is indeed important, particularly with regard to the AI Act, to have clear governance in terms of the use of artificial intelligence, and in particular to be able to explain how data is used as well as the mechanisms that lead to a result," continues Marvin Sant. Our solution makes it possible to do this, assuring organizations that they are acting in compliance with applicable regulations. »

An assistant for every need

Devana has a simple interface that makes it easy for users and those responsible for configuring it and deploying new wizards to learn. "It offers the possibility of selecting the company's databases with which the models will be able to interact in real time in order to meet well-identified needs," adds Marvin Sant.

In this way, virtual assistants can be configured according to the user's needs, based on pre-trained models, by limiting access to certain data. An HR manager may, for example, have a virtual assistant who supports him in the performance of his missions, based on information relating to personnel management, without being able to access customer data. Another assistant, with other levels of authorization, can be deployed to support the legal department. "The solution can also be easily integrated with knowledge bases commonly used in companies, for a Sharepoint, a OneDrive or a Google Drive or be interfaced with a CRM or an ERP, depending on what you want to do Marvin Sant explains.

A simple and intuitive solution

Today, Fujitsu is supporting Luxembourg players in the adoption of this extremely powerful tool, which allows its users to quickly take advantage of the advantages of artificial intelligence, by facilitating the adoption of AI in a serious way among all employees. "The solution is relatively simple and intuitive. It allows you to quickly become familiar with these virtual agents. It's even possible to connect multiple agents together, in order to meet more complex needs while limiting access to certain source data for each, explains Steve Heggen. It is easily configurable, like gears that can be assembled and set in motion to meet well-identified needs. The solution adapts to the organization, the specific constraints of a given market, and regulatory requirements while helping to improve the user experience. »

A solution implemented quickly

The solution is deployed within 3 to 6 months, with an initial exploration phase, intended to fully understand the tool, then an experimentation stage, to validate the first use cases. The final phase is enterprise-wide deployment, to enable the entire organization to become more efficient.