About QDNA
QDNA is an independent research notebook on sovereign artificial intelligence platforms, on-premises or hybrid. It publishes analyses to help choose hardware, open models and architecture, and to frame compliance. It is kept on a personal, non-professional basis: QDNA is not a company, sells no services and neither installs nor operates any platform.
About this blog
Aurélien Ambert, CISSP.
Architect and pre-sales consultant for artificial intelligence and inference platforms.
He also works for an integrator that designs, sells, integrates and operates artificial intelligence platforms, an activity separate from QDNA: it is in that role that he designs and operates local LLM inference platforms for production, from deploying vLLM and NVIDIA NIM engines on GPU farms to tracking throughput in tokens per second and latency. His background spans eighteen years in infrastructure, networking and security, with a specialisation in cloud-native security.
Certifications: CISSP (ISC2) and Microsoft Certified Azure Solutions Architect Expert.
Areas of expertise:
- open-weight models
- vLLM and NVIDIA NIM
- Kubernetes
- multi-agent orchestration
- data sovereignty
How the analyses are built
The analyses follow a five-step method, from the need to planning day-to-day operation:
- Start from the need and the organisation's use cases.
- Recommend the hardware suited to the organisation's profile.
- Choose interchangeable open models, served through a single gateway.
- Define multi-agent orchestration and memory.
- Plan operations in compliance with GDPR, the objectives of the HDS certification and SecNumCloud.
Implementing and operating a platform remain with each organisation's teams or the provider of its choice, in its own datacentres or in managed ones. QDNA only publishes analyses.
Projects
The public sites published by QDNA illustrate the architectures described in the analyses: gironde.qdna.fr, mmr.qdna.fr and apprendre-ia.qdna.fr.