Our committments
Sovereignty and data protection
We work with institutions and companies that possess sensitive data. From the design of our software to our operating and hosting conditions, we ensure the security of the documents entrusted to us as part of a processing project.
Hosting in France and Germany
All of our data storage and processing infrastructure is hosted in Europe. This choice ensures that all of our customers' data remains subject to European regulations.
Full traceability of the processing chain
By controlling every step of the technical chain, we ensure the protection and integrity of our customers' data. Our Arkindex software allows us to accurately identify the processing applied to documents, thereby controlling data usage.
Open Standards
We strongly believe in the lasting benefits open source. We show our dedication to its values of collaboration and transparency by making our tools and knowledge freely available.
Giving back to the community
We make extensive use of open source technologies in our daily workflows and are committed to giving back to the community by publishing our own productions.
Designing a sustainable service
The majority of our projects are led by the public sector, which requires long-term service continuity. We ensure that our software and services can be maintained beyond the commercial collaboration with our customers through open-source.
Sharing our expertise
Open source is also a way for our developers to demonstrate the quality of their work, receive constructive feedback from their peers, and contribute to the growth of artificial intelligence applications in the heritage ecosystem.
Responsable AI
Our teams strive to minimize the carbon footprint of our models and infrastructure, while ensuring that data is used ethically.
Developing small, specialised models
Instead of relying on large, generalist Language Learning Models (LLMs), we focus on developing smaller, specialised models. This targeted approach ensures that the models are not only more efficient at their specific tasks, but also consume less energy due to their optimised size and scope.
Reuse of pre-trained models
Our commitment to open source allows us to use pre-trained models for decoding or fine-tuning tasks. This practice of reusing existing models means that we can reduce the duration and intensity of the training processes, thereby saving energy. By avoiding the need to train models from scratch, we significantly reduce computational resources and energy consumption.
Efficient softwares and code
The greatest energy consumption in web software often lies not in its operation, but in its development and maintenance. We chose Python and PyTorch for their efficiency, as they offer significant energy and time savings throughout the software lifecycle.
Quality of life at work
We promote an inclusive and flexible work culture, where everyone can thrive.
Remote work and flexible schedules allow our teams to maintain a healthy work–life balance while achieving excellence.