The abstract is read at the publisher; the summary is JClub's.
EcoToxFred, a knowledge-graph-grounded conversational agent for European surface-water chemical monitoring and hazard data, selected the right tool in at least 92% of runs and reproduced reference queries in 74%.
- Why it matters: Integrated interpretation of monitoring, ecotoxicity, and derived risk indicators is hard for non-experts because the needed data are rarely accessible without database skills.
- What they did: The system combines an LLM with a Neo4j knowledge graph in a retrieval-augmented generation setup, translating questions into graph queries and returning traceable outputs. It was tested on 25 benchmark questions, each asked three times, using GPT-4.1 and GPT-5 with tool, execution, and RAGAS goal scoring.
- The result: Tool choice reached at least 92% and reference-query reproduction was 74% with both models, while goal accuracy was 52% for GPT-4.1 and 63% for GPT-5. The approach enables traceable ecotoxicology data exploration and is presented as transferable to other ecological data infrastructures.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.