Tag
This paper presents CourseChat, an on-premises multi-course RAG tutoring system for undergraduate business education, sharing six isolated courses across two edge AI hosts running Ollama-served local LLMs. Model bake-offs show that 8B/12B/7B models pass classroom speed gates while larger ones fail, supporting a hardware-software co-design approach rather than assuming bigger models are optimal.
This paper investigates agentic retrieval that combines LLM reasoning with dense retrievers in a ReAct loop for complex search tasks, showing an 8.7-point nDCG@10 improvement over standard retrieval and strong generalization on ViDoRe v3 and BRIGHT leaderboards, but at a substantial latency and token cost of 107.4 seconds and 764K input tokens per query.
本文研究 agentic RAG 评测中评估预算在问题数、搜索轨迹和重复读取之间的分配策略，通过 HotpotQA 和 MuSiQue 上的检索-反馈对比，发现更广的问题覆盖在固定 token 预算下能显著降低标准误差，并提出了预算分配的预测模型。
该推文盘点了 10 个用于让 AI 智能体为其输出提供可验证证据的开源 GitHub 项目，涵盖科学文献检索（OpenScholar、PaperQA2）、文档解析（Docling）、网络爬取（Crawl4AI）、RAG（RAGFlow、GraphRAG、LightRAG）、深度研究（Open Deep Research）以及事实核查与评估（DeepEval、Ragas），并重点展示了 Allen AI 的 OpenScholar 代码库。
该论文研究了在单语检索增强生成（RAG）中推理语言与查询/检索文档语言对齐的作用，构建了一个基于虚构桌游世界《The DarkEye》的德语 RAG 问答测试集，发现对齐推理语言能提升表现，但仅能达到模型原生英语推理的水平。
This paper applies split-conformal claim filtering to multi-hop RAG, showing that at a 95% conformal target, the fraction of responses with fully supported claims rises to 95.80%-97.20% across HotpotQA, Natural Questions, and TriviaQA using Llama 3.1 8B and GPT-4o-mini — but with strong abstention and low claim retention, highlighting that reliability gains trade off against answer coverage.
PathAnchor 是一个基于路径结构证据工作区的科学推理系统，通过 Material–Sensor–Signal–System 轨迹组织证据，在 120 个柔性传感器问题上以 82.6% 得分领先六个基线系统，并显著提升来源召回率与引用完整性。
SeKondBrain AI Labs introduces the Concept Lifecycle Model and Concept-Grounded Attention, evaluating graph-injected attention, temporal versioning, and epistemic status on QA benchmarks. The results support explicit temporal validity and epistemic status while showing that several graph-attention mechanisms fail controlled comparisons with disabled-mechanism baselines.
论文提出 SourceLearn 框架，帮助 AI 智能体在反复使用同一知识来源时持续积累可复用的源特定理解，而非每次都进行全新的检索任务。结果显示该方法在 15 个设置中 13 个取得最佳结果，最高较 Hybrid RAG 提升 22.6 分，且对检索缺失更鲁棒。
介绍 Jev 如何作为 RAG 管道中的“裁判”，在检索与生成之间评估候选段落的相关性、可回答性和潜在提示注入，并附上基于 Comet Opik 的开源 Jev-as-a-Judge 项目。
This paper formalizes SFIA-based skill and responsibility-level extraction from free text, evaluating five strategies (lexical baseline, dense retrieval with LLM reranking, zero-shot schema-constrained LLM, single-agent agentic RAG, and a three-agent retriever-matcher-Verifier crew), finding that explicit level prediction is key and multi-agent crews add latency without accuracy gains.
Sieve and Sage 是一个两阶段 RALM 拒答框架，用轻量级 Sieve 模块在调用昂贵 LLM (Sage) 之前过滤干扰性检索证据，并区分不可答与被干扰两种状态，相比单阶段基线最高提升 69.4%p 准确率、55.2%p Macro-F1，并实现 1.99 倍加速。
The paper introduces ARCagent, an adaptive retrieval calibration agent for clinical question answering in ME/CFS, where diagnostic guidelines actively contradict each other. It contributes a conflict-aware retrieval re-ranking pipeline and an LLM-as-Judge benchmark, achieving 95.3% accuracy and outperforming base LLMs.
This paper introduces LLM-Guided Graph Pruning (LGP), a framework that uses LLM reasoning to refine graph-based ANN indices (e.g., HNSW, DiskANN) by pruning low-value neighbors, addressing the geometry–semantic mismatch in how these indices are built versus evaluated, and improving end-to-end retrieval over vanilla greedy search and LLM reranking.
EvoDuet introduces a bilevel optimization framework that co-evolves solutions and web search queries with fixed LLM parameters, using a retrieval gate to decide when to fetch new documents versus reusing stored ones. Across 21 tasks it boosts OpenEvolve's normalized discovery gain notably (e.g., 61.3% to 82.3% with Gemini-3.8-Flash), surpassing prior best scores on eight tasks and generalizing across different evolutionary search scaffolds.
The article discusses the challenge of distinguishing between retrieval and generation failures in RAG systems and explores practical methods to measure each component independently.
研究人员推出名为 PageIndex 的全新 RAG 方案，无需向量数据库、嵌入或分块，而是构建树状索引让大模型像人类读书一样检索，在 FinanceBench 上以 98.7% 的准确率超越所有向量 RAG 方法，且完全免费开源。
This paper proposes TRACE, a fine-tuning framework for Retrieval-Augmented Generation (RAG) that uses multi-agent debate traces and answer completeness regularization to handle knowledge conflicts, improving robustness against misleading context and reducing incomplete answers.
Researchers introduced PageIndex, a novel RAG approach that eliminates reliance on vector databases, embeddings, and chunking, achieving 98.7% accuracy on FinanceBench and outperforming existing methods while being free and open source.
This paper introduces LogicTree-RAG, a logic tree-guided retrieval-augmented generation framework for automating long-form patent drafting, enhancing content quality and token efficiency compared to existing methods.