evidoria

← Back to browse

Good practice Imported

data.gouv.fr's Albert-Powered AI Metadata Enrichment and Natural-Language Catalogue Access

France · Paris · See the France profile · See the Paris profile

Evidence: Descriptive / self-reported Top 74% 47/100 · Ask Evidence Copilot about this practice

France's open-data portal uses Albert, the state's open LLM, to auto-draft dataset descriptions and suggest metadata tags, and exposes its catalogue via an MCP server for natural-language AI queries — though Etalab warns AI suggestions may be wrong and need human review.

data.gouv.fr's Albert-Powered AI Metadata Enrichment and Natural-Language Catalogue Access

Details

Maturity
Pilot
Promoter
Etalab (DINUM, French Prime Minister's Office)
Period
2025–present
Keywords
open data, generative AI, metadata, government digital services

Context

Etalab, the French digital administration agency inside DINUM (Direction interministérielle du numérique), has added generative-AI features to data.gouv.fr, the national open-data portal, as part of the platform's 2025 evolution roadmap.

Activities

The features rely on Albert, a French state-hosted open large language model: it auto-generates short descriptions for datasets whose producers haven't written one, and suggests keywords and tags drawn from existing metadata to improve catalogue completeness. The portal was also connected to an MCP (Model Context Protocol) server, letting external AI assistants such as Claude, ChatGPT or Mistral query the dataset catalogue directly in natural language.

Conclusions

Etalab has not published adoption or accuracy statistics for these features, and describes the work as experimental and iterative rather than a finished, measured product — its own documentation states plainly that 'suggestions from LLMs may be incomplete, approximate, or wrong' and that human validation is required before publishing. Independent French tech-and-data coverage (samsa.fr) has flagged a related security caveat: fake MCP servers impersonating data.gouv.fr have already appeared online, underlining that opening a government data catalogue to LLM querying carries new risks alongside its convenience gains.

Implementation
Indicative cost
Low (< €50k) — No budget disclosed; conservative low estimate as the feature reuses existing state LLM (Albert) infrastructure rather than new model development.
Time to results
Short (< 1 year)
Staffing & skills
Delivered by Etalab (DINUM), the French interministerial digital agency, building on Albert, the state's shared open LLM

Conditions for success

  • Reuses existing state AI infrastructure (Albert LLM) rather than building new models
  • Requires human validation before publishing AI-suggested metadata, per Etalab's own documentation

Common failure modes

  • Etalab's own documentation states LLM suggestions 'may be incomplete, approximate, or wrong'
  • Independent press (samsa.fr) has identified fake MCP servers impersonating data.gouv.fr, a security risk from opening the catalogue to LLM querying
  • No adoption or accuracy statistics have been published for the AI features

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful