MDIA: A Multi-Agent Diagnostic Intelligence Pipeline on HealthBench Professional

Roberto Cruz, David Rey-Blanco · 2026

Read MDIA: A Multi-Agent Diagnostic Intelligence Pipeline on HealthBench Professional by Roberto Cruz, David Rey-Blanco (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-24699.txt.

Description

Roberto Cruz and David Rey-Blanco argue that clinical benchmark gains come less from prompt tricks than from system design. Their MDIA pipeline routes cases through a seven-node graph of specialties on the full HealthBench Professional benchmark, without fine-tuning. Specialty routing, multi-turn context, drug-state safety gating, filtered search, length-aware synthesis, and engine reliability drive the lift. It scores 3.72 points above ChatGPT for Clinicians on the benchmark. Yet swapping the grader shifts the score noticeably, so the authors urge evaluation across several independent graders.

Catalog record

CREATORRoberto Cruz, David Rey-Blanco
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69788
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.24699

Browse the full library · About the Tower of Babel