DrugClaw and DrugAudit: A Primary-Source-Grounded Agent and Authority-Aware Benchmark for Drug-Information Question Answering
Researchers have developed DrugClaw, a multi-agent system designed for accurate drug-information question answering. This system utilizes a reflection-driven workflow to query drug registries and provide answers grounded in primary regulatory or peer-reviewed records. To evaluate its performance, they also created DrugAudit, a benchmark comprising 3,772 items, which assesses source match, snippet overlap, and citation faithfulness. DrugClaw demonstrated superior performance across all metrics on DrugAudit and related medical question-answering datasets. AI
IMPACT Enhances accuracy and trustworthiness in drug information retrieval, crucial for clinical decision-making and regulatory compliance.