DILA · Services du Premier ministre: Building a Sovereign AI Platform to Accelerate RAG & Agentic AI Delivery
A sovereign, reusable AI foundation to accelerate the delivery of RAG and agentic applications.
ROLE
AI Product & Delivery Leadership | Sovereign GenAI (9-person team)
CLIENT
DILA, Services du Premier ministre
FOCUS
Sovereign AI · Agentic AI · RAG · Platform engineering · AI governance
YEAR
2026
Only public or client-approved information is shared here: no internal architecture, documents, data or confidential figures.
01 / 05
The challenge
Public institutions need to deploy generative AI while maintaining strict requirements for data sovereignty, security, traceability and regulatory compliance.
However, building each AI application independently creates significant duplication: document ingestion, embeddings, retrieval, orchestration, security controls, evaluation and observability often have to be rebuilt for every project.
The objective was to move beyond isolated AI proofs of concept and establish a sovereign, reusable AI foundation capable of accelerating the development of production-ready RAG and agentic applications.
02 / 05
My contribution
I led a 9-person team on the product and delivery approach of a sovereign GenAI platform, with a strong focus on modular architecture, component reuse and scalable AI delivery.
03 / 05
The approach
Sovereign AI architectureDefine a platform approach compatible with strict hosting, security and data-sovereignty requirements, including SecNumCloud considerations.
Reusable AI building blocksDefine shared capabilities for document ingestion, chunking, embeddings, hybrid retrieval, reranking, prompt management and LLM integration.
Agentic orchestrationStructure reusable workflows for tool calling, multi-step reasoning and agent-based task execution.
RAG factoryStandardize the creation of document-based AI assistants using configurable retrieval pipelines rather than project-specific implementations.
Evaluation frameworkIntroduce systematic RAG quality measurements, including faithfulness, answer relevancy, context precision and context recall.
Governance by designEmbed access management, traceability, auditability and sovereignty requirements into reusable platform components.
Delivery accelerationEstablish repeatable implementation patterns to reduce development effort and improve consistency across AI projects.
04 / 05
Impact
From isolated AI projects to a reusable platform. The approach shifted the development model from building each application independently to assembling standardized, governed AI components.
New RAG applications could reuse shared document processing, retrieval, evaluation and security capabilities. Agentic applications could build on existing orchestration patterns and tool integrations.
This reduced duplicated engineering work, accelerated experimentation and established a more consistent approach to AI quality and governance.
70%
Faster content creation
−60%
RAG application delivery time, from 10 weeks to 4 weeks
75%
of common technical capabilities reused
0.94
RAG faithfulness
−30%
cost per AI request, through pipeline optimization and reuse
Agentic workflow prototyping: −50% implementation time
Answer relevancy: 0.91
Context precision: 0.89
Context recall: 0.92
Evaluation coverage: 85% of priority RAG scenarios
Figures confirmed with the project team. Evaluation method shared on request.
05 / 05
What I learned
The real value of an enterprise AI platform is not only the intelligence it provides, but how quickly and reliably it lets an organization build its next AI application. Treating RAG pipelines, agentic workflows, evaluation and governance as reusable platform capabilities turns sovereign AI into a scalable engineering model rather than a collection of disconnected experiments.
