📄 中文摘要
PageIndex 是一种无向量的 RAG 框架，在金融文档的准确率上达到了 98.7%，而 GPT-4o 仅为 31%。该框架通过构建层次树索引来保留文档结构，避免了将文档分块成向量的过程，从而支持大型语言模型的推理能力。这种方法展示了文档检索的理想工作方式，解决了传统 RAG 系统在处理特定问题时常常返回语义相似但内容错误的段落的问题。
PageIndex 与向量 RAG：为何基于推理的检索实现了 98.7% 的准确率（而非 31%）
📄 English Summary
PageIndex vs Vector RAG: Why Reasoning-Based Retrieval Achieves 98.7% Accuracy (Not 31%)
PageIndex is a vectorless RAG framework that achieved an impressive 98.7% accuracy on financial documents, compared to only 31% for GPT-4o. Instead of chunking documents into vectors, it constructs a hierarchical tree index that preserves the document structure and supports reasoning capabilities of large language models (LLMs). This approach illustrates how document retrieval should have functioned from the beginning, addressing the common issue where traditional RAG systems return semantically similar but contextually incorrect sections when queried about specific information.