Roquette leverages Azure OpenAI Service in the age of generative AI

From technological excitement to the creation of an in-house chatbot.

RoqGPT was born of one observation: employees' unbridled enthusiasm for ChatGPT. "The application quickly caught on with our employees. Within a few months, at least 500 of them had tested the online version," explains Pierre-Louis Bescond, Head of Data & Advanced Analytics. This enthusiasm prompted the company to consider how to integrate this technology without compromising the high level of data confidentiality it has to comply with. "We wanted to avoid at all costs the disclosure or use of the questions asked via this tool to re-train future models,” adds Mathieu Klimczak, Lead DevOps in the Data & Advanced Analytics department.

The group then decided to develop its own generative AI chatbot: RoqGPT. It had to offer the same conversational applications as ChatGPT, plus several additional functionalities. As for the prompts and responses generated by this new tool, they would only be accessible to end-users, with a reinforced level of confidentiality.

At the same time, Microsoft launched Azure OpenAI Service. This was a godsend for the Group, which seized the opportunity. "Our entire infrastructure was already based on Azure, so it seemed obvious to bank on these technologies once again," says Pierre-Louis Bescond. The maturity of the product was also enticing: "With Microsoft and Azure OpenAI, we have access to the most powerful Large Language Models on the market today, which means we can offer our employees an internal tool as powerful as ChatGPT. What's more, we were able to integrate these technologies into our stack in just a few hours.” Other major advantages include the high level of security that is native to Azure, and the fact that it takes sustainable development issues into account. "Thanks to Azure's API system, we only use the inference of our AI models, not their training. This enables us to reconcile sustainability, economic challenges and innovation."

Implementation through experimentation and acculturation

It took Roquette just four hours to access its first Azure OpenAI Service resource. This was followed by the creation of a test group, and the development of a platform to simplify access to the large language models (LLMs) inherent in generative AI. A few months later, the chatbot was tested with a panel of 300 employees, before being extended to all Group entities worldwide. Rapid and efficient, the deployment was deliberately carried out in infrastructure as code ("IaC") mode, in order to create an agile and reproducible technological solution and environment. "We want to be able to deploy on-demand in a single command line," emphasizes Pierre-Louis Bescond.

At the same time, onboarding is becoming more structured, because while RoqGPT is intuitive to use, "it's important to acculturate employees to this technology, whether in terms of good security practices or developing relevant prompts," analyzes Mathieu Klimczak. Seminars, workshops, video tutorials... Everything is done to ensure that everyone fully embraces the tool, in a "self-service" approach. "Our innovations are automatically accompanied by an increase in the autonomy and skills of our teams. This is one of the founding principles of Roquette's digital transformation," adds Pierre-Louis Bescond. "Artificial intelligence, reinforced by the arrival of Generative AI, enables us to continue creating value through innovative digital solutions, whether in a "Self-Service" mode for all Roquette Group employees or through specialized solutions."

Embracing AI: the example of the HR and Audit departments

Many departments have adopted the application, including Human Resources: "RoqGPT helps us write letters, job descriptions, and short articles," explains Océane Robin, Head of HR Shared Services Center France. HR also uses the chatbot to reformat data, produce meeting minutes and feed inventiveness in terms of team animation and cohesion.

On the Audit side, RoqGPT could well become an ally for auditors. "The Internal Audit department carries out around twenty assignments a year in various areas of the company," comments Laurence Delecroix, Corporate Internal Auditor. The guideline each of these audits has to follow comprises a number of key stages. These include the mission preparation phase (risk identification, data analysis, audit work programs), and Laurence Delecroix recently tested generative AI for one of her missions. "RoqGPT provides relevant audit trails and risk identification. It enables us to process this information in a secure environment, which is essential in our line of business."

Making room for productivity, creativity, and people

In just a short period of time, the number of users has passed the 1,000 mark, and millions of tokens—"digital assets"—are generated daily by the application. "These figures are a perfect illustration of the commitment to the project and its positive impact on our teams’ work," notes Mathieu Klimczak.

More generally, two to four hours of productivity have been gained every week since RoqGPT went live. Low value-added tasks have been relegated to technology, leaving more room for people and creativity. "Generative AI is a source of opportunity and transformation. This gives us more scope to focus on the human side," stresses Océane Robin. “It releases us from the usual constraints and boosts our imagination," adds Pierre-Louis Bescond.

What's more, RoqGPT strengthens the commitment of employees, who are proud and motivated by the idea of working with such a tool, just as it arouses the curiosity of potential candidates: "The fact that Roquette is positioning itself on this type of subject increases our attractiveness. It's a great lever for our employer brand."

Towards new technological horizons

Thanks to these initial successes, RoqGPT's future looks promising. What's next? Adding image generation functionality, integrating the latest OpenAI models, querying a Corpus of proprietary documents on-demand, improving the overall user experience and developing more detailed monitoring of the solution.

"AI is a constantly evolving field, and it's impossible to know what tomorrow's technologies will look like. It's up to us to build a strategy that is both robust and scalable over time," concludes Pierre-Louis Bescond.