arXiv:2511.15994v1 Announce Type: cross 
Abstract: Access to the right evidence does not guarantee that large language models (LLMs) will reason with it correctly. This gap between retrieval and reasoning is especially concerning in clinical settings, where outputs must align with structured protocols. We study this gap using Written Exposure Therapy (WET) guidelines as a testbed. In evaluating model responses to curated clinician-vetted questions, we find that errors persist even when authoritative passages are provided. To address this, we propose an evaluation framework that measures accuracy, consistency, and fidelity of reasoning. Our results highlight both the potential and the risks: retrieval-augmented generation (RAG) can constrain outputs, but safe deployment requires assessing reasoning as rigorously as retrieval.

تدرس الدراسة الفجوة بين استرجاع الأدلة والتفكير في نماذج اللغة الكبيرة (LLMs) في السياقات السريرية، باستخدام إرشادات العلاج بالتعرض الكتابي (WET) كحقل اختبار. على الرغم من توفير معلومات موثوقة، تستمر الأخطاء في استجابات النماذج. يتم اقتراح إطار تقييم يقيس الدقة والاتساق والأمانة في التفكير، مما يبرز الحاجة إلى تقييم صارم جنبًا إلى جنب مع عمليات الاسترجاع.

El estudio examina la brecha entre la recuperación de evidencia y el razonamiento en modelos de lenguaje grandes (LLMs) en entornos clínicos, utilizando específicamente las directrices de la Terapia de Exposición Escrita (WET). A pesar de proporcionar información autorizada, persisten errores en las respuestas de los modelos. Se propone un marco de evaluación para medir la precisión, consistencia y fidelidad del razonamiento, enfatizando la necesidad de una evaluación rigurosa junto con los procesos de recuperación.

L'étude examine l'écart entre la récupération d'informations et le raisonnement dans les grands modèles de langage (LLMs) dans des contextes cliniques, en utilisant spécifiquement les directives de la thérapie d'exposition écrite (WET). Malgré la fourniture d'informations autorisées, des erreurs dans les réponses des modèles persistent. Un cadre d'évaluation est proposé pour mesurer l'exactitude, la cohérence et la fidélité du raisonnement, soulignant la nécessité d'une évaluation rigoureuse en parallèle des processus de récupération.

The study examines the gap between evidence retrieval and reasoning in large language models (LLMs) within clinical settings, specifically using Written Exposure Therapy (WET) guidelines. Despite providing authoritative information, errors in model responses persist. An evaluation framework is proposed to measure reasoning accuracy, consistency, and fidelity, emphasizing the need for rigorous assessment alongside retrieval processes.

CARE-RAG - Clinical Assessment and Reasoning in RAG

arXiv:2511.16681v1 Announce Type: new 
Abstract: Retrieval-Augmented Generation (RAG) systems have become a dominant approach to augment large language models (LLMs) with external knowledge. However, existing vector database (VecDB) retrieval pipelines rely on flat or single-resolution indexing structures, which cannot adapt to the varying semantic granularity required by diverse user queries. This limitation leads to suboptimal trade-offs between retrieval speed and contextual relevance.
  To address this, we propose \textbf{Semantic Pyramid Indexing (SPI)}, a novel multi-resolution vector indexing framework that introduces query-adaptive resolution control for RAG in VecDBs. Unlike existing hierarchical methods that require offline tuning or separate model training, SPI constructs a semantic pyramid over document embeddings and dynamically selects the optimal resolution level per query through a lightweight classifier. This adaptive approach enables progressive retrieval from coarse-to-fine representations, significantly accelerating search while maintaining semantic coverage.
  We implement SPI as a plugin for both FAISS and Qdrant backends and evaluate it across multiple RAG tasks including MS MARCO, Natural Questions, and multimodal retrieval benchmarks. SPI achieves up to \textbf{5.7$\times$} retrieval speedup and \textbf{1.8$\times$} memory efficiency gain while improving end-to-end QA F1 scores by up to \textbf{2.5 points} compared to strong baselines. Our theoretical analysis provides guarantees on retrieval quality and latency bounds, while extensive ablation studies validate the contribution of each component. The framework's compatibility with existing VecDB infrastructures makes it readily deployable in production RAG systems. Code is availabe at \href{https://github.com/FastLM/SPI_VecDB}{https://github.com/FastLM/SPI\_VecDB}.

تم اقتراح إطار عمل جديد يسمى Semantic Pyramid Indexing (SPI) لتحسين أنظمة الجيل المعزز بالاسترجاع (RAG) من خلال السماح بفهرسة متجهات متعددة الدقة في قواعد بيانات المتجهات (VecDBs). تتناول هذه المقاربة المبتكرة قيود خطوط أنابيب الاسترجاع الحالية التي تعتمد على هياكل الفهرسة المسطحة، والتي تواجه صعوبات مع التباين الدلالي في استفسارات المستخدمين.

Se ha propuesto un nuevo marco llamado Semantic Pyramid Indexing (SPI) para mejorar los sistemas de Generación Aumentada por Recuperación (RAG) al permitir la indexación vectorial de múltiples resoluciones en bases de datos vectoriales (VecDBs). Este enfoque innovador aborda las limitaciones de los pipelines de recuperación existentes que dependen de estructuras de indexación planas, que tienen dificultades con la granularidad semántica variable en las consultas de los usuarios.

Un nouveau cadre appelé Semantic Pyramid Indexing (SPI) a été proposé pour améliorer les systèmes de génération augmentée par récupération (RAG) en permettant un indexage vectoriel multi-résolution dans les bases de données vectorielles (VecDBs). Cette approche innovante répond aux limitations des pipelines de récupération existants qui reposent sur des structures d'indexation plates, qui peinent à gérer la granularité sémantique variable dans les requêtes des utilisateurs.

A new framework called Semantic Pyramid Indexing (SPI) has been proposed to enhance Retrieval-Augmented Generation (RAG) systems by allowing for multi-resolution vector indexing in vector databases (VecDBs). This innovative approach addresses the limitations of existing retrieval pipelines that rely on flat indexing structures, which struggle with varying semantic granularity in user queries.

CARE-RAG - Clinical Assessment and Reasoning in RAG

Was this article worth reading? Share it

Augmeta

HubRE AI

Aidx.ai