DENIA: Multilingual AI Microservice Orchestration Practice at France Médias Monde

The DENIA system developed by France Médias Monde is an AI microservice orchestrator based on the open-source Mistral 7B, supporting transcription, translation, summarization, and semantic classification in over 15 languages, serving 2000 users and 61 editorial projects. This article will analyze this practice from aspects such as background, architecture, application evidence, and technical details.

In the global media environment, multilingual content processing is a core challenge for international news agencies. France Médias Monde needs to handle a large amount of cross-language news daily; the traditional manual process is costly and lacks real-time performance, which promoted the development of the DENIA system.

DENIA adopts a microservice architecture, with core designs including: built on the Mistral 7B open-source model to avoid complete reliance on commercial APIs; a routing architecture supporting seamless switching between open-source models and commercial APIs; coverage of over 15 languages; and scalable deployment to serve a large number of users and projects.

DENIA has served 61 editorial projects and approximately 2000 users. Its core functional implementations include: integrating the OpenAI Whisper API for multilingual speech transcription; a neural network translation module ensuring semantic accuracy of content; automatic summarization improving editorial screening efficiency; and semantic classification supporting content recommendation and retrieval.

At the technical level, the project uses FastAPI to build RESTful APIs and combines Spacy for NLP; it emphasizes the data annotation process, supporting custom NER model training to identify entities in the news domain; and provides an Azure cloud deployment solution to achieve seamless migration from development to production.

The practice of DENIA provides important references for AI applications in the media industry: balancing open-source and commercial solutions ensures data sovereignty and performance; domain customization improves application effectiveness; progressive deployment reduces risks; and human-machine collaboration enhances overall efficiency.

In the future, DENIA-like systems will play a role in more language scenarios. It is recommended that developers and organizations refer to the ideas of this project such as open-source architecture and domain customization; its open-source project also provides valuable practical experience for multilingual AI applications.