BNP Paribas deploys an internal LLM-as-a-Service platform hosted in its own data centers

Overview

BNP Paribas designed and deployed an internal LLM as a Service platform, operated by the Group's IT teams, giving entities standardized, secure access to open-source models and Mistral AI models on shared GPU infrastructure hosted in the Group's data centers. It was deployed in an initial test phase across entities such as Hello bank! and the Group's general inspection, with internal assistants, document generation and document information retrieval use cases in production or experimentation.

The challenge

BNP Paribas wanted to facilitate the development and deployment of generative AI use cases across its businesses and functions, giving Group entities unified and secure access to large language models.

The solution

BNP Paribas designed and deployed an internal LLM as a Service platform, operated by the Group's IT teams. Each entity accesses a selection of language models tailored to its needs through a standardized and secure interface, and the models can be integrated directly into tools or processes. The platform runs on shared GPU infrastructure hosted in the Group's data centers and offers open-source models and models from Mistral AI; models trained on internal datasets are planned to follow ("soon"), according to specific needs.

Reported business value

According to BNP Paribas, the shared infrastructure gives high processing power while sharing and limiting costs, and facilitates scaling up generative AI use cases within a common framework with shared technical and security standards. The source gives no quantified results.