{"paper":{"title":"SlideAgent: Hierarchical Agentic Framework for Multi-Page Visual Document Understanding","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"SARA hybrid RAG keeps some text passages and compresses the rest into vectors to raise answer quality under fixed token limits.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Rachneet Kaur, Srijan Kumar, Sumitra Ganesh, Yiqiao Jin, Zhen Zeng","submitted_at":"2025-10-30T15:41:15Z","abstract_excerpt":"Multi-page visual documents such as manuals, brochures, presentations, and posters convey key information through layout, colors, icons, and cross-slide references. While multimodal large language models (MLLMs) offer opportunities in document understanding, current systems struggle with complex, multi-page visual documents, particularly in fine-grained reasoning over elements and pages. We introduce SlideAgent, a versatile agentic framework for understanding multi-modal, multi-page, and multi-layout documents, especially slide decks. SlideAgent employs specialized agents and decomposes reason"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SARA consistently improves answer relevance (+17.71), answer correctness (+13.72), and semantic similarity (+15.53) across 9 datasets and 5 open-source LLMs spanning 3 model families.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that semantic compression vectors remain interpretable enough to enable effective iterative evidence reranking while preserving the benefits of the retained text snippets, as described in the hybrid framework proposal.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SARA combines natural-language snippets with semantic compression vectors in RAG to improve answer relevance, correctness, and similarity on 9 datasets across 5 LLMs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SARA hybrid RAG keeps some text passages and compresses the rest into vectors to raise answer quality under fixed token limits.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d57441aa9a6bac0090e4e811c7960241315b21b3986795885f6de96f54403361"},"source":{"id":"2510.26615","kind":"arxiv","version":4},"verdict":{"id":"41fbec24-e345-40b6-94d6-f0c0ddde7f93","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T03:00:22.110768Z","strongest_claim":"SARA consistently improves answer relevance (+17.71), answer correctness (+13.72), and semantic similarity (+15.53) across 9 datasets and 5 open-source LLMs spanning 3 model families.","one_line_summary":"SARA combines natural-language snippets with semantic compression vectors in RAG to improve answer relevance, correctness, and similarity on 9 datasets across 5 LLMs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that semantic compression vectors remain interpretable enough to enable effective iterative evidence reranking while preserving the benefits of the retained text snippets, as described in the hybrid framework proposal.","pith_extraction_headline":"SARA hybrid RAG keeps some text passages and compresses the rest into vectors to raise answer quality under fixed token limits."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2510.26615/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"37c6d44df0ddf2c0ade5df197498eafa52a79be3b2b4971b53d063e6a4c7458c"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}