CohortFlow: A Steerable Clinical Cohort Curation Agent with Localized Semantic Correction
arXiv:2610.36100v1 Announce Type: new Abstract: Language model agents for clinical cohort curation should accommodate evolving eligibility criteria without repeatedly processing unchanged requirements. We introduce CohortFlow, a steerable agent that represents natural language corrections as localized updates to a structured cohort specification and reconciles them with a persistent execution graph. Explicitly recorded semantic dependencies enable affected computations to be selectively invalidated while preserving valid prior results. On 100 scenarios derived from clinical-trial eligibility c
阅读原文 · arXiv 人工智能本文由 SoccerDream AI 自动聚合索引，内容版权归原作者与 arXiv 人工智能 所有。
本站仅作技术动态聚合，不构成任何形式的建议。
本站仅作技术动态聚合，不构成任何形式的建议。