Arbor: A Framework for Reliable Navigation of Critical Conversation Flows

Researchers introduce Arbor, a framework that decomposes large language model decision-making into specialized node-level tasks for critical applications like healthcare triage. The system improves accuracy by 29.4 percentage points while reducing latency by 57.1% and costs by 14.4x compared to single-prompt approaches.

• →Arbor framework addresses LLM struggles with structured workflows in high-stakes domains by breaking decision trees into specialized tasks.
• →The system uses DAG-based orchestration to dynamically retrieve and evaluate only relevant decision paths at runtime.
• →Testing across 10 foundation models showed 29.4 percentage point improvement in turn accuracy over baseline approaches.
• →Per-turn latency decreased by 57.1% while achieving 14.4x reduction in operational costs.
• →Architectural decomposition enables smaller models to match or exceed performance of larger models using traditional methods.