VectifyAI/PageIndex
PageIndex is a vectorless, reasoning-based RAG system that builds a hierarchical tree index from long documents for agentic, context-aware retrieval, simulating human expert navigation.
Core Features
Detailed Introduction
Traditional vector-based RAG often struggles with long professional documents due to reliance on semantic similarity over true relevance, leading to inaccuracies. PageIndex addresses this by introducing a vectorless, reasoning-based RAG system. Inspired by AlphaGo, it constructs a hierarchical tree index of documents and employs LLMs to reason over this index, mimicking how human experts navigate complex information. This approach enables agentic, context-aware retrieval without vector databases or artificial chunking, offering superior explainability and achieving state-of-the-art accuracy in professional document analysis.