Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2608.11033.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Paper introduces XstrAI, a multi-agent framework that converts SHAP feature-attribution outputs into audience-tailored narratives for patients, clinicians, and data scientists. It uses immutable shared evidence and three specialized LLM agents (planning, linguistic realization, validation) with a bounded revision loop, evaluated on diabetes and stroke risk prediction against 11 baselines.

The immutable-evidence-plus-validator pattern is a reusable multi-agent design where faithfulness to source data is enforced structurally rather than prompted.

Cited by

No citations on record.