TRACE (Transparent Research and Claim Evidence) - Unlocking Flexible Knowledge Graphs Through Atomic Claims Architecture

Short description of the innovation study

AI answers are only as trustworthy as the evidence behind them. Today's systems pull up text fragments without knowing whether sources agree or contradict each other, which is a problem in high-stakes fields like journalism, climate finance and policy analysis. TRACE breaks documents into individual, traceable claims and automatically shows which ones support or contradict each other, so every answer comes with its evidence and sources. We will fine-tune compact AI models for this, run on EuroHPC supercomputing, and test them on real challenges, including: checking whether government and corporate climate commitments are grounded in IPCC science. Expected results: open models, a graph of 100,000+ claims, a query tool for evidence-backed answers.

Organisations involved: 

Main SME: The Newsroom