Learn2RAG

Retrieval augmented generation for SMEs

Project description:

Data-driven companies usually have to manually select a combination from the large number of available RAG approaches, software modules and basic AI models that can fit their specific application. Added to this are the various data sources that need to be integrated into the RAG architecture. The key challenge when using RAG in practice lies in the informed assembly of adequate RAG pipelines based on their intended use. This is where Learn2RAG comes in by automating the learning of company-specific RAG pipelines and their deployment.

The aims of the project are
i) the development and evaluation of a data-efficient, supervised machine learning method for the learning of company-specific RAG pipelines based on their intended application and
ii) the industry-compliant provision of the learnt RAG pipelines.
This will enable German organisations (especially SMEs enterprises without AI expertise) to obtain correct, up-to-date and explainable answers from large language models without significant investment. Through the consortium's existing network, at least 150 companies will be reached and enabled to test and use RAG already during the project period.

Consortium:

University of Paderborn;
IfDT - Institute for Digital Technologies gGmbH;
USU GmbH;
German Red Cross Rhineland-Palatinate Regional Association;
Fraunhofer Society for the Promotion of Applied Research, registered association IEM

Duration:

February 2025 - January 2028

Budget:

Total funding: € 2.8 million
Funding amount: € 2.3 million