Press Release: vector8’s Recognition at the Trustworthy AI Summit Hosted at EDF Lab
Vector8 was recognized at the Trustworthy AI Summit for its Smart Router Collaborative Research Project with BNP Paribas, which developed a dynamic routing solution that automatically selects the optimal large language model for each query to achieve significant cost savings (up to 10–15x) while optimizing accuracy and sustainability.
Vector8 and BNP Paribas presented their Smart Router collaborative research project at the Trustworthy AI Summit, showcasing an intelligent model-routing system that selects the optimal LLM for each query. The approach delivered up to 10–15x cost savings while maintaining or improving accuracy, and highlights a scalable path toward industrial, responsible AI. The project continues to expand with a V2 focused on broader use cases, energy efficiency, and sustainable AI deployment.
Our teams at vector8 and BNP Paribas, represented by Solveig Bachellery Head of Innovation Exposition at BNP Paribas and by Gilles Mazars and Samuel JAÏS for vector8, were honored to share our latest milestone regarding our Smart Router Collaborative Research Project, at this Europe’s flagship event for industrial and responsible AI, along 250+ participants – scientists, industrial leaders and regulators – gathered to exchange on the challenges and opportunities of trustworthy AI.
With the explosion of LLMs, choosing the right model for each query has become a real challenge. Our dynamic smart routing solution builds an intelligent layer on top of “LLM gateways” platforms, automatically selecting the best model for every request—optimising for cost and accuracy in this paper, but can be extended to energy and compliance.
Presented under the scientific poster section showcasing the latest breakthrough in AI, our architecture leverages domain-specific datasets and real-world leaderboards to ensure the right model is chosen every time.
The results? We’ve demonstrated up to 10–15x cost savings while maintaining or even improving correctness - outperforming the best single model approach (in the context of the models pool and datasets chosen for the experiment).
Our journey continues as we develop V2, expanding to more use cases and refining our strategy for even greater impact. This is a step forward in industrialising AI deployment at scale, empowering teams to evaluate and specialise models efficiently and driving sustainability with energy and carbon footprint control.
We are grateful our work has sparked such a strong interest among the distinguished crowd of Researchers and AI professionnals, and were pleased by the consensus encountered by our approach towards building industrial AI deployments capacities.
This event also marked the official launch of the European Trustworthy AI Association - led by - Air Liquide, NAVAL GROUP, Safran, IRTSystemX and Thales - created to offer a reference methodology and tools to promote the adoption of responsible AI and enable industrial-scale.
.svg)
