Sopra Steria Model Training Augmented Generation

Our customers in the industry sector can optimize their processes by leveraging AI models through different use cases :​

• predictive maintenance
• referentials update leveraging activity footage
• employee empowerment in their daily tasks (Is the vehicule ready and can I start it ? Has the vehicule met an obstacle, of which type and how to handle it ?)

The AI models involved in these use cases need a large amount of data/images, which can be very long or even impossible to acquire in the field.​ Indeed, highly reliable systems will rarely generate data associated with failures, which our customers seek to prevent.​ Generating this data (especially images) can be tedious, even with image generation tools, because these tools must be configured/parameterized to generate a large amount of data with particular attention to borderline cases: extreme weather, specific lighting, specific obstacles and varied combinations of these parameters.​

LLMs and GenAI can improve this data or image generation process by generating the scenarios to be processed by the image generation tools and some of the assets to be used (where generation takes precedence over precision).​ The images thus generated must be evaluated in relation to the reference images (quality measurement) and, if necessary, corrected (domain adaptation).​ To effectively meet these business needs, our teams are implementing and continuing to develop for these customers a methodology and assets that takes advantage of LLMs, image and data generation tools, quality measurement and correction algorithms co-developed as part of Confiance.ai as well as AWS services.​

This offer heavily relies on AWS services such as : Amazon Sagemaker, Amazon Sagemaker Pipelines, Amazon Bedrock, Amazon EC2, Amazon ECS, AWS Lambda, Amazon S3.

Several customers from the Transport, Aeroline, Retail & Defense sectors are already trusting us and improving their industrial use cases with our solutions !