DEVOPS REX 2025 | Industrialising a sovereign LLMaaS: scalability, security... and AI that helps to create itself

DEVOPS REX returns to the Cité des Sciences in 2025!

Can we create a sovereign generative AI service with the help of AI itself?

Read our exclusive feedback: how we industrialised a sovereign LLMaaS, hosted on a SecNumCloud IaaSdedicated to sensitive customers - government sectors, healthcare, etc.

The challenges of a reliable and secure LLM

We'll take a look behind the scenes at how a product goes into production Reliable, scalable, measurable and secure LLM (Large Language Model). What's original about it? The AI itself was involved in its creation!

Concrete challenges addressed:

• Heterogeneous GPU support (Nvidia A100, L40S, H200, Apple M4 ...) with dynamic scheduling according to workloads
• Token billingthanks to a distributed counting system integrated into the loadbalancers
• Detailed performance measurementlatency, tokens per second, efficiency per model
• Intelligent load distribution (4 active routers, full supervision)
• SecNumCloud integrationIAM, audit, isolation, sovereign storage

AI, the driving force behind DevSecOps innovation

AI has contributed to its own industrialisation :

• Generation of infra-as-code (Kubernetes manifests, adaptive proxies)
• Automatic production of dashboards and supervision scripts
• Help with diagnosing production errors
• Automated routing and resource allocation

Why take part?

This exchange will be an opportunity to :

• Understanding the challenges of governance and security sovereign generative AI
• Discover the tools, successes and limitations encountered on an AI project in production
• Explore transferable good practice AI projects in companies
• Discussing culture DevSecOpsthe observability and sovereignty of the cloud