Tobias Schreieder

EVIDENZ: Evidence-Based Text Generation with Large Language Models

The EVIDENZ project aims to enhance the trustworthiness of large language models (LLMs) in scientific application scenarios. The focus is on developing methods for evidence-based text generation, in which generated content is systematically linked to traceable and verifiable references. This enables users to independently validate statements generated by LLMs using scientific sources.

EVIDENZ addresses key challenges of current LLMs, which often produce linguistically convincing responses but fail to provide citations or reference only weakly supporting sources. To address this problem, the project investigates both non-parametric approaches that integrate external scientific publications during generation and parametric methods that enable generated content to be traced back to a model’s training data.

A central component of the project is the development of a novel benchmark for evidence-based scientific text generation. The benchmark is designed to evaluate both the quality of generated responses and the correctness and relevance of their citations, enabling systematic comparisons between different approaches. Building on this benchmark, the project develops retrieval-augmented generation (RAG) methods that retrieve and integrate relevant scientific evidence while considering the citation function, as well as scalable attribution methods for future scholarly AI systems.

The collaboration with Springer Nature provides access to large-scale scientific publications and metadata, enabling the developed methods to be evaluated on realistic scholarly communication tasks.

Project page: EVIDENZ

AICOM: Corporate Knowledge Graph Assistant

AICOM develops trustworthy enterprise AI assistants by combining LLMs with corporate knowledge graphs. Instead of relying solely on model memory, the system retrieves evidence from structured and unstructured organizational knowledge, enabling answers that are traceable, verifiable, and grounded in company-specific information.

The project investigates knowledge graph construction and maintenance, retrieval-augmented generation over enterprise knowledge, and methods for assessing answer quality through evidence coverage, consistency, and source traceability. The resulting assistant will support knowledge-intensive workflows such as documentation, reporting, and decision support while increasing transparency and trustworthiness.

Project page: AICOM

Powered by Jekyll and Minimal Light theme.