Sovereign AI: French Field Report
summary
A municipal project built a local Retrieval-Augmented Generation (RAG) system called Marmotia to let city agents query vectorized internal documents, comparing a vendor-hosted RAG deployment in Shawinigan with an open-source Docker stack deployment in Chambéry. The prototype revealed practical issues: local LLMs are slower with limited token throughput causing responsiveness problems, PDF export can break vectorization, and operational needs like data normalization, a governance charter, and organizational change proved more important than model choice. Continued experimentation of the RAG prototype is planned.
AI-written summary · about
chapters (15)
- 0:00Team and project
- 1:08Shawinigan partner
- 3:16Agile methodology
- 5:03AI sovereignty legal
- 6:06Field immersion sprint
- 8:14RAG solutions applications
- 10:46Success limits feedback
- 12:56User needs challenges
- 14:43Data interoperability
- 16:27Transformation error rights
- 18:54Public sector innovation
- 19:57AI sovereignty vision
- 20:57AI education acculturation
- 21:59Frugality territorial cooperation
- 22:41Conclusion
related talks
51mDeveloping a Custom RAG System
46mBypassing the Multimodal Tax: Hybrid RAG, SQL RRF & UI Telemetry
50mModular RAG Architectures with Java and Spring AI
31mBuild an Agentic RAG system using Langchain, Ollama and Milvus
40mRolling out Enterprise AI: Tools, Insights, & Team Empowerment
41mFrom RAGs to riches: Build an AI document interrogation app in 30 mins
16mOpenRAG: An open-source stack for RAG
31mAdvanced Enterprise RAG Systems