Adaptive Prompt Routing and Hallucination Detection for Enhanced Multilingual Open-Source LLMs

Short description of the innovation study

This innovation study addresses the critical challenge of bridging the gap between open-source generative AI and production-grade applications in European industries. It aims to develop state-of-the-art multilingual large language models (LLMs) that significantly reduce computational overhead by training only the most pertinent parameters, thereby lowering infrastructure costs while enhancing language diversity. A novel LLM routing system is proposed to dynamically select optimal models, achieving up to an 85% reduction in inference expenses. Concurrently, the study integrates advanced explainability techniques for real-time hallucination detection, ensuring transparent and reliable outputs. Leveraging high-performance computing and cutting-edge training methodologies, the project aspires to establish a robust, regulation-compliant AI framework that reinforces digital sovereignty and fosters competitive innovation across Europe.

Organisations involved: 

End User: Seedbox Ventures GmbH